Louisiana State UniversityLSU Digital Commons
LSU Doctoral Dissertations Graduate School
2015
Optimized Aperiodic Multilayer Structures forAbsorbers and Thermal EmittersChristopher H. GranierLouisiana State University and Agricultural and Mechanical College
Follow this and additional works at: https://digitalcommons.lsu.edu/gradschool_dissertations
Part of the Physical Sciences and Mathematics Commons
This Dissertation is brought to you for free and open access by the Graduate School at LSU Digital Commons. It has been accepted for inclusion inLSU Doctoral Dissertations by an authorized graduate school editor of LSU Digital Commons. For more information, please [email protected].
Recommended CitationGranier, Christopher H., "Optimized Aperiodic Multilayer Structures for Absorbers and Thermal Emitters" (2015). LSU DoctoralDissertations. 3259.https://digitalcommons.lsu.edu/gradschool_dissertations/3259
OPTIMIZED APERIODIC MULTILAYER STRUCTURESFOR ABSORBERS AND THERMAL EMITTERS
A Dissertation
Submitted to the Graduate Faculty of theLouisiana State University and
Agricultural and Mechanical Collegein partial fulfillment of the
requirements for the degree ofDoctor of Philosophy
in
The Department of Physics and Astronomy
byChristopher H. Granier
M.S.,Louisiana State University, 2014August 2015
Acknowledgments
I would like to express my gratitude to my co-advisors Dr. Jonathan P. Dowling and Dr.
Georgios Veronis for their constant guidance and unwavering support throughout part of
my undergraduate and my entire graduate-school career. Jon’s enthusiasm, innovativeness,
and imagination, all well-grounded in the principles of meaningful research and a deep un-
derstanding of physics and its applications allow him to generate interesting research topics
which will undoubtedly lead to discoveries useful in everyday life. Dr. Veronis’s painstakingly
methodical approach to research coupled with his extreme attention to detail and countless
hours of consultation were invaluable during my graduate career. Dr. Veronis provided his
assistance and experience in debugging computer codes as well as his uncanny ability to
anticipate and carefully craft counter arguments to would-be reviewers. Words cannot serve
as a medium to adequately express my gratitude to both Jon and Dr. Veronis.
I would also like to thank Dr. Lee, who was always willing to help with any questions
I might have had throughout my graduate career and Dr. Corbitt, who has provided an
experimental insight into my work in nanophotonics.
Over the past five years, I have had the opportunity to work with two incredibly bright
undergraduates. Francis Afzal assisted with some of the computational work associated with
Chapter 2 and some of the work presented in Chapter 3 over the course of two summers
spent at LSU as a REU student. Simon Lorenzo recently started working with us and has
assisted in streamlining the computer codes as well as research associated with the Dual-
Angle Absorber found in Chapter 3 and the lighhtbulb work in Chapter 4. I want to thank
both of them for their contributions to the work as well as their friendship and support.
Special thanks to my peers, past and present, in the QST group, who provided helpful
discussion and good company: Ashkan Balouchi, Bryan Gard, Manish Gupt, Kaushik Se-
shadreesan, Todd Moulder, Siddhartha Das, Bhaskar Roy Bardhan, Robinjeet Singh, Zhihao
ii
Xiao, Kebei Jiang, Nick Lanning, and Jonny Olson. I would especially like to thank the
person who will be moving forward with some of the work presented here and someone who
has become a close friend, Chenglong You.
I also want to thank my entire family; each person has a special place in my life and
has played a pivotal role in making me who I am today. Finally, and most importantly, I
would like to thank my mother and grandmother, whose love and self-sacrifice have known
no bounds over the past 28 years. Words cannot express my gratitude.
iii
Table of Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
Chapter 1. Introduction and Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Computational Electricity and Magnetism . . . . . . . . . . . . . . . . . . . 2
1.1.1 Transfer Matrix Formalism . . . . . . . . . . . . . . . . . . . . . . . . 31.2 Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2.1 Blackbody Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . 161.3 Kirchoff’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.1 Solar Thermophotovoltaics for Power Generation . . . . . . . . . . . 191.4 Discussion of Optimization Algorithms . . . . . . . . . . . . . . . . . . . . . 23
Chapter 2. Optimized Aperiodic Highly Directional Narrowband Infrared Emitters . . 262.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.1 Fitness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3.1 Aperiodic Emitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312.3.2 Wien Wavelength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382.3.3 Aperiodic Structure as the Emitting Portion of a Detector . . . . . . 40
2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Chapter 3. Optimized Aperiodic Multilayer Structures for Use as Narrow-Angular Ab-sorbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3.1 Tungsten-Silicon and Tungsten-Silica Structures . . . . . . . . . . . . 523.3.2 Silica-Silicon Structures . . . . . . . . . . . . . . . . . . . . . . . . . 593.3.3 Multiple Material Structures . . . . . . . . . . . . . . . . . . . . . . . 653.3.4 Bichromatic Absorber . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Chapter 4. How to Build a Better Lightbulb — Exploiting Optimization Algorithms toDecrease Incandescent Lightbulb Power Consumption by Nearly 50 % . . . . . . . . . . . . . 734.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 734.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 744.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Chapter 5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
iv
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Appendix: Permission to Use Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
v
Abstract
In this dissertation, we begin with a brief introduction to nanophotonics. In particular,
we will focus on the theory of computational electricity and magnetism, specifically, the
method used for this research, the transfer-matrix method. We will also provide discussions
of the principles which motivate the analysis of the multilayer structures: Planck’s blackbody
distribution, Kirchoff’s law, and thermodynamics. Finally, we will provide a brief discussion
of computer-based optimization algorithms, specifically NLOPT (a library which provides
computational packages for nonlinear optimization) and genetic algorithms. This background
information is necessary to understand discussions which occur in later chapters.
Chapter 2 will discuss multilayer structures which are optimized by a genetic algorithm
to provide both narrowband and narrow-angle thermal emission for selected wavelengths
in the infrared wavelength range. This chapter will compare the performance of aperiodic
multilayer structures with more widely researched periodic multilayer structures. It will also
provide a detailed analysis of how this emission profile is achieved via an electromagnetic field
analysis. Finally, Chapter 2 will present a concrete application for such optimized structures
as a carbon monoxide detector via absorption spectroscopy.
Chapter 3 will focus on the development of aperiodic multilayer structures for use as
narrow-angle absorbers. Both the layer thicknesses and materials used are optimized by a
genetic optimization algorithm coupled to a transfer matrix code. We find that utilizing
silicon and silica above a thick tungsten substrate provides a structure capable of near unity
absorptance at a single wavelength. Finally we show that structures with almost perfect
absorptance at multiple wavelengths for normally incident light can be achieved.
Chapter 4 presents optimized aperiodic structures for use as broadband, broad-angle ther-
mal emitters. These structures are capable of increasing the thermal emittance by nearly a
factor of two when compared to bulk tungsten. We utilize a hybrid optimization algorithm
vi
coupled to a transfer matrix code to maximize the power emitted in the visible wavelength
range in the normal direction. Chapter 4 also discusses the mechanisms present which al-
low these structures to possess properties, which could lead to a decrease in incandescent
lightbulb power consumption by nearly 50%.
vii
Chapter 1Introduction and Theory
Photonic crystals are a widely known field of ongoing research which involves fabricating
or simulating structures in one, two, or three dimensions for some specific use from solar
thermophotovoltaics to antireflective coatings. Photonic crystals, by nature, exhibit period-
icity; photonic nanostructures, however, do not. These nanostructures may be fabricated or
computationally explored in one, two, or three dimensions and seek to affect the motion
of photons in such a way that a specific use is achieved – high absorptivity, high reflectiv-
ity, angular selectivity, etcetera. Some of the most common applications kept in mind when
designing these structures are: coherent emitters, solar themophotovoltaics, and filters. Re-
search in these areas is done largely computationally. Electromagnetic codes implementing
Transfer Matrix, FDFD (finite difference frequency domain), FDTD [1] (finite difference time
domain), and plane wave expansion methods are among some of the most frequently used.
FIGURE 1.1: Schematic [2] of photonic crystals of one, two, and three dimensions. Figure 1.1shows that a one dimensional photonic crystal only has one degree of freedom; while the otherdirections are considered infinite or periodic (depending upon the boundary conditions used).Two dimensional photonic crystals have two variable widths (a checkerboard); while, threedimensional photonic crystals have three variable sides (such as the sides of stacked cubes).The source of this figure is Photonic Crystals: Molding the Flow of Light, by Joannopoulos,Meade, and Winn.
1
As such, nanophotonics is an interdisciplinary field, which combines thermodynamics, elec-
tricity and magnetism, and computational methods to achieve structures suitable for many
possible applications. These structures may vary in complexity, and, due to the difficulty and
expense in fabricating and modeling more complex structures (two- and three-dimensional
structures), we choose to focus on one-dimensional, aperiodic structures. We computation-
ally model these structures and implement the transfer matrix formalism, which is coupled
to either a genetic or hybrid optimization algorithm.
1.1 Computational Electricity and Magnetism
To fully investigate the properties of these nanophotonic structures, we will first begin with
a detailed analysis of their electromagnetic properties. Starting with Maxwell’s equations,
we will develop a transfer-matrix formalism which we can use to analyze a one dimensional,
aperiodic structure’s absorption, transmission, and reflection properties.
Electromagnetic waves in free space obey Maxwell’s equations:
∇ · E = 0, (1.1)
∇ ·B = 0, (1.2)
∇× E = −∂B
∂t, (1.3)
∇×B = µ0ε0∂E
∂t, (1.4)
where ε0 and µ0 are the free space permittivity and permeability, respectively. Additionally
we identify the propagation velocity, c is the speed of light such that µ0ε0 = c−2. By taking
the curl of Equation 1.3 and 1.4, making use of the vector identity,
∇×∇×A = ∇(∇ ·A)−∇2A, (1.5)
we find,
∇(∇ · E)−∇2E = −∂∇×B
∂t= −µ0ε0
∂2E
∂t2, (1.6)
∇(∇ ·B)−∇2B = µ0ε0∂∇× E
∂t= −µ0ε0
∂2B
∂t2, (1.7)
2
as well as Equations 1.1 and 1.2, we arrive at the free-space equation for electromagnetic
waves:
∇2E− 1
c2∂2E
∂2t= 0, (1.8)
∇2B− 1
c2∂2B
∂2t= 0. (1.9)
1.1.1 Transfer Matrix Formalism
We consider light incident on an aperiodic, one-dimensional, multilayer structure composed
of multiple materials with different indexes of refraction and thicknesses. Here, we restrict
ourselves to materials where µ = µ0, but ε 6= ε0, and is, in fact, the relative permittivity
of the material. Additionally, we consider structures which are infinite in both the x- and
y-directions. Under these conditions, Maxwell’s equations change slightly. Most notably, we
FIGURE 1.2: Light is incident from vacuum on an aperiodic n-layer structure com-posed of multiple materials with indexes of refraction N1, N2, ...., Nn and thicknessesD1, D2, D3, ..., Dn.
replace ε0 with ε and c is no longer the speed of light in vacuum, but the speed of light in
3
the ith layer is given by ci = c/Ni, such that:
∇ ·D = ρ, (1.10)
∇ ·B = 0, (1.11)
∇× E = −∂B
∂t, (1.12)
∇×H = J +∂D
∂t, (1.13)
where J is the electric current density and ρ is the free charge density present [3, 4]. In the
cases we consider here, there is no free charge or electric current density, so Equations 1.10
and 1.13 simplify, yielding:
∇ ·D = 0, (1.14)
∇×H =∂D
∂t, (1.15)
where D = εE.
To analyze the properties of such a structure, we must first consider the solutions to the
electric field wave equation. By examining Maxwell’s equations in the frequency domain and
assuming an e−jωt harmonic time dependence for all field quantities, and, to avoid confusion,
j =√−1, we can write Maxwell’s equations in the following way:
∇ ·D = ρ, (1.16)
∇ ·B = 0, (1.17)
∇× E = jωB, (1.18)
∇×B = J− jωD. (1.19)
A full solution is then just the superposition of solutions found for each individual frequency.
Solutions will then have the form:
Aej(ki·r), (1.20)
4
where, for example, the wave vector kx,i in the ith layer is given by:
kx,i =2πNi sin θi
λ0, (1.21)
Ni is the index of refraction in the ith layer, θi is the angle of propagation in the ith layer and
λ0 is the wavelength in free-space. Additionally, we can now suppress the time dependence of
the solution. More specifically, in the presence of a multilayer structure which reflects some
portion of incident radiation, the field to the left of the multilayer structure is given by:
E(x, y, z) = 1ej(kx,ix+kz,iz) + re−j(kx,ix−kz,iz), (1.22)
where we have normalized the incident field amplitude to be 1 and r is the reflection coef-
ficient of the structure. As such, we have shown how electromagnetic radiation propagates
in free space, and by extension inside the layers of our multilayer structure. Now, we must
discern the effects that the interfaces between the layers have on the electromagnetic fields.
First, we begin by examining light incident at an angle θi as in Equation 1.21. To achieve
this, we take a closer look at Figure 1.3 and examine the boundary conditions.
We consider monochromatic light propagating in the +z-direction and incident upon the
boundary between two materials which possess indexes of refraction N0 and N1. We are
interested in the angle of propagation of light reflected from the boundary, θr, and the
angle of propagation of light which is transmitted into the second medium, θt. We consider
harmonic plane waves which result from the interaction with the interface of the form:
Aej(ki·r), (1.23)
Ae−j(kr·r), (1.24)
Aej(kt·r), (1.25)
for each the incident, reflected, and transmitted wave. For there to be a constant relation
which exists for all points of the interface, the phase, φ = k · r of each the incident, re-
flected, and transmitted waves must be equal at the interface. Thus, we apply the boundary
5
FIGURE 1.3: Light is incident at an angle θi on an interface between two materials withindexes of refraction N0 and N1 respectively.
conditions at z = 0, the interface between N0 and N1 such that:
ki · r|z=0 = kr · r|z=0 = kt · r|z=0. (1.26)
For z = 0, this leads to the relation:
kx,ix = kx,rx = kx,yx, (1.27)
which must be true for all values of x; therefore, we have:
kx,i = kx,r = kx,t, (1.28)
Finally, with k0 = 2πN0
λ0and k1 = 2πN1
λ0we can rewrite these using their constituent angles:
kx,i = k0 sin θi, kx,r = k0 sin θr, kx,t = k1 sin θt, kz,i = k0 cos θi, kz,r = k0 cos θr,
kz,t = k1 cos θt, and, finally, ky,i = ky,r = ky,t = 0 and with kx,i = kx,r, we recover: θi = θr.
Making use of the relation derived from Equation 1.28, kx,i = kx,t, we recover Snell’s Law:
N0 sin θi = N1 sin θt. (1.29)
6
While Snell’s law determines the direction of propagation for both the reflected and trans-
mitted waves at the interface between two materials with difference indexes of refraction,
it gives no information about the amplitude of the two waves. To establish a relationship
between the incident, transmitted, and reflected amplitudes, we must further investigate the
boundary conditions which apply. To achieve this, we will, once again, make use of Maxwell’s
equations, this time, in integral form, adjusted for electric current density, J = 0 and free
charge, ρ = 0 : ∮D · dS = 0, (1.30)∮B · dS = 0, (1.31)∮
E · dl = − ∂
∂t
∫S
B · dS, (1.32)∮H · dl =
∂
∂t
∫S
D · dS. (1.33)
First, we will apply the boundary conditions; then, we will apply these boundary conditions
to both TE (Transverse-Electric), also known as s-Polarized, and TM (Transverse-Magnetic),
also known as p-Polarized, light.
We first consider the two closed surface integrals, and, as depicted in Figure 1.4 (a), we
assume that the interface is surrounded by a cylinder. From Equations 1.30 and 1.31, we
find that in region N0, dS = −qdA and in region N1, dS = qdA, so these equations yield:
DN0 · q = DN1 · q, (1.34)
BN0 · q = BN1 · q. (1.35)
This implies that the components of B and D in the normal direction are continuous across
the boundary between N0 and N1. Now, we consider the line integrals in Equations 1.32 and
1.33, which, as depicted in Figure 1.4 (b) involve constructing a loop across the boundary
between N0 and N1. We note that as we let dh→ 0, the fluxes of both B and H across the
7
FIGURE 1.4: (a) We enclose the boundary between N0 and N1 with a cylinder of height dh,surface area dA whose sides are perpendicular to the surface of the interface, and q is theunit vector from medium N0 to medium N1. (b) We enclose the boundary between N0 andN1 with a loop of height dh and surface area dA whose short sides are perpendicular to thesurface of the interface and q is the unit vector from medium N0 to medium N1. Here, p isthe unit vector which is parallel to the surface of the interface between N0 and N1.
boundary go to zero; therefore,
EN0 · p = EN1 · p, (1.36)
HN0 · p = HN1 · p. (1.37)
From this, we can surmise that E and H’s tangential components are continuous across
the boundary between N0 and N1. Now, we will apply this to the s-Polarization of light as
depicted in Figure 1.5.
We begin by considering incident light at an angle θi to the surface of an interface between
two materials:
Ei = Eiej(−ki·r)y, (1.38)
8
FIGURE 1.5: For Transverse Electric (s) polarized light, the electric field is perpendicular tothe plane of incidence, and the magnetic field is parallel.
where, here,
ki =ωN0
c(sin θix + cos θiz). (1.39)
Similar relations exist for Bi, Er, Br, Et, and Bt. Due to the fact that there is no normal
component of the electric field for transverse electric polarized light, the condition of conti-
nuity of the D field is satisfied by construction. We are left with the boundary conditions
that the normal (z-direction) component of B and the tangential (x-direction) component
of H be continuous:
N0 sin θi(Ei + Er) = N1 sin θtEt, (1.40)
N0 cos θi(Ei − Er) = N1 cos θtEt. (1.41)
Here, we have suppressed the µ which would come from the magnetic field equations due to
the fact that we only consider materials where µ = µ0. Thus, we can normalize the incident
9
field such that |Ei| = 1, rearrange Equations 1.40 and 1.41, and find that the reflection and
transmission coefficients are:
rTE =N0 cos θi −N1 cos θtN0 cos θi +N1 cos θt
, (1.42)
tTE =2N0 cos θi
N0 cos θi +N1 cos θt. (1.43)
We then can calculate the reflection and transmission coefficients for transverse magnetic
polarized light by a similar process and by taking into account the difference in the electric
and magnetic field directions as depicted in Figure 1.6.
FIGURE 1.6: For Transverse Magnetic (p) polarized light, the electric field is parallel to theplane of incidence, and the magnetic field is perpendicular.
The continuity equations are:
cos θi(Ei − Er) = cos θtEt, (1.44)
10
Bi + Br = Bt, (1.45)
which then, by relating E and B in the following way, B =N
cE, where N is the index
of refraction in the first or second medium, leads to the transverse magnetic reflection and
transmission coefficients,
rTM =−N0 cos θt +N1 cos θiN0 cos θt +N1 cos θi
, (1.46)
tTM =2N0 cos θi
N0 cos θt +N1 cos θi. (1.47)
As expected, for θi = 0, we recover:
rTE = rTM =N1 −N0
N0 +N1
, (1.48)
tTE = tTM =2N0
N0 +N1
, (1.49)
under the Fresnel equation symmetry relations [5], which allow us to equate the following
coefficients:
rN0,N1 = −rN1,N0 , (1.50)
tN0,N1tN1,N0 + rN0,N1rN1,N0 = 1. (1.51)
Now, we must consider how to approach both the boundary conditions and free propaga-
tion in tandem. For this, we return to Figure 1.2 and Equations 1.20 and 1.21 to formulate
the transfer matrix method. We consider the field to the left (and in principle in the ith layer)
of the multilayer structure for z < 0, which now, is composed of both a forward-propagating
(in the positive z-direction of amplitude Ei+) and a reflected wave (traveling in the negative
z-direction of amplitude Ei+) of the ith layer.
E(x, y, z) = Ei+e
j(kz,iz)ej(kx,ix) + Ei−e−j(kz,iz)ej(kx,ix), (1.52)
where k is as defined in Equation 1.21. Henceforth, we will suppress the variation in the
x-direction, as it only serves to keep track of phase in that direction. Now, we can discern
11
a matrix which will propagate the changes in the electromagnetic field (phase accumulation
and attenuation of the field) across a layer. That is, we will compose a matrix equation for
Ei+1+ and Ei+1
− from Ei+ and Ei
+, and we have dropped the vector form of the electric field
due to the fact that the fields which are not in the direction of propagation do not contribute
to this analysis: Ei+1+
Ei+1−
=
ejφ 0
0 e−jφ
Ei
+
Ei−
, (1.53)
where φ =2πNiDi cos θi
λ, and, again, Ni and θi are the index of refraction and angle of
propagation in the ith layer, Di is the thickness of the ith layer. That is, we account for the
phase change and, in principle, loss in the following way:
Ei+1+ = ejφEi
+, (1.54)
Ei+1− = e−jφEi
−, (1.55)
which means that the forward (rightward) propagating wave accumulates a phase ejφ, and
the backward (leftward) propagating wave gains a phase e−jφ.
Finally, let us consider writing the boundary conditions in matrix form. Due to internal
reflection and transmission, we could, in principle, have both forward and backward propa-
gating waves for both Ei+1− and Ei+1
+ , so, we desire a matrix which will operate in the following
manner:
Ei+1+ = ti,i+1E
i+ + ri+1,iE
i+1− , (1.56)
Ei− = ri,i+1E
i+ + ti+1,iE
i+1− . (1.57)
Now, Equations 1.56 and 1.57 form a system of equations which we can solve and isolate
Ei+ and Ei
− from Ei+1+ and Ei+1
− , which will allow us to combine the boundary conditions
with the propagation matrix shown in Equation 1.53. Solving the system of equations and
12
using the symmetry relations of the Fresnel coefficients, found in Equations 1.50 and 1.51,
we can accomplish this: Ei+
Ei−
=
1ti,i+1
ri,i+1
ti,i+1
ri,i+1
ti,i+1
1ti,i+1
Ei+1
+
Ei+1−
. (1.58)
Thus, we have established a transfer-matrix formalism for the boundary conditions, and need
only to combine it with the propagation through a layer found in Equation 1.53; therefore, we
define δi,i+1, as the boundary condition matrix for the boundary between the i and i+1 layer,
and, by solving Equation 1.53 for the fields in the ith layer, we can recover the propagation
matrix for the ith layer, Πi :
δi,i+1 =
1ti,i+1
ri,i+1
ti,i+1
ri,i+1
ti,i+1
1ti,i+1
, (1.59)
Πi =
e−jφ 0
0 ejφ
. (1.60)
Thus, we have successfully related the fields on the left and right hand sides of the n-layer
structure in Figure 1.2: E0+
E0−
= δ0,1
n−1∏i=1
Πiδi,i+1Πn
En+
0
, (1.61)
where En− is identically zero because there is no incident field from the right side of the
structure. This allows us to extract the overall reflection and transmission coefficients for
the structure. By rewriting the product which corresponds to the boundary conditions and
propagation matrices, we find: E0+
E0−
=
M1,1 M1,2
M2,1 M2,2
En
+
0
. (1.62)
Solving the system of equations for the reflection, r, and transmission, t, coefficients of the
structure yields:
E0−
E0+
= r =M2,1
M1,1
, (1.63)
13
En+
E0+
= t =1
M1,1
. (1.64)
In principle, if there were a semi-infinite substrate (as there is in many calculations presented
in this dissertation), the calculation would end with δn−1,n and the transmission coefficient
would be identically zero. We can then calculate the power reflected, R, transmitted, T , and
absorbed, A, by the structure:
R = |r|2 =M2,1
M1,1
M∗2,1
M∗1,1
, (1.65)
T = |t|2 =1
M1,1
1
M∗1,1
, (1.66)
and, by conservation of energy,
ATE/TM(λ, θ) = 1−RTE/TM(λ, θ)− TTE/TM(λ, θ), (1.67)
where the dependence of the absorptance, reflectance, and transmittance upon polarization,
wavelength, and angle of incidence is explicitly shown.
We can go through a similar process, which will be outlined here, to reach a transfer-
matrix formalism for the E and H fields which will give insight as to the physical mechanisms
which drive the absorption and emission of the structures. The reflection and transmission
coefficients for the structure, gotten by the previously outlined method, are necessary to
compute the fields. We find that the electric and magnetic fields, as a function of z through
the first layer of thickness D1, again, as depicted in Figure 1.2, are: E(z)
Z0H(z)
=
cos(2πNizλ
)− jNi
sin(2πNizλ
)−jNi sin
(2πNizλ
)cos(2πNizλ
) E(z = z0)
Z0H(z = z0)
, (1.68)
where E(z = z0) is the field calculated by the method shown in Equations 1.61-1.64 and
Z0 is the free-space impedance. Thus, we are able to extract the field at each position z as
the field moves from left to right in the structure. It is important to note that at the end of
each layer, we multiply by an additional propagation matrix after accounting for the phase
accumulated in that layer by setting z = Di, where Di is the thickness of the ith layer. We
14
can also calculate the power as a function of z within the structure, and, as a result, locate
areas of high and low loss within the structure.
Starting with the Poynting vector, which computes the rate of energy transfer per unit
area,
S = E×H. (1.69)
We can rewrite Equation 1.69 in a more useful form, assuming a sinusoidal field and com-
puting the time-averaged value of the Poynting vector S, by considering the integral over a
time T :
< S >=1
T
∫ T
0
S(t) dt, (1.70)
which then yields the time-averaged Poynting vector:
< S >=1
2Re(E×H∗). (1.71)
Thus, we have established methods for calculating the absorption, transmission, and reflec-
tion coefficients for an n-layered one-dimensional structure. We have also outlined a method
for extracting the exact fields within the layers of the structure and shown that the power
(and therefore, the loss) can be calculated.
1.2 Thermodynamics
Many of the applications of nanophotonic structures are motivated by the laws of thermo-
dynamics. These structures have clear applications as absorbing devices, useful for, among
other things, light filters and solar energy conversion. We must investigate the process of
the emission of light from heated objects, which will be useful in this end. While we have
only explored a method for calculating the absorptance of multilayer structures, we make
use of Kirchoff’s law to equate absorption and emittance under the condition of thermal
equilibrium. Using Kirchoff’s law, in conjunction with Planck’s theory of Blackbody radia-
tion is necessary to explore the usefulness of these aperiodic multilayer structures as thermal
15
emitters, which have a wide variety of applications, including spectroscopic detectors and
replacements for incandescent lightbulbs.
1.2.1 Blackbody Radiation
The theory of Blackbody radiation [6] is well-known. Under the condition of thermal equilib-
rium, we consider a cube with sides of length L. In each dimension, we note that the allowed
momenta and wavelengths are:
λ =2L
nand p =
hn
2L, (1.72)
where n is a positive integer used to label the specific oscillatory mode, h is Planck’s constant,
and c is the speed of light. Here, the energy associated with any photon is
E(n) =hcn
2L. (1.73)
Thus, we consider these quantities in three-dimensions:
E(n) = c√p2x + p2y + p2z =
hc
2L
√n2x + n2
y + n2z, (1.74)
and use Equation 1.72 again, where we consider n as the magnitude of the nx, ny, nz three
vector. Now, we make use of the partition function [7] in statistical mechanics, which is
defined:
Z(β) =∞∑n=0
e−βE(n) =e−
βE(n)2
1− e−βE(n), (1.75)
where β = 1kBT
, kB is the Boltzmann constant, and T is the temperature. Finally, we can
calculate the expected energy in each mode:
< E >= −∂ln[Z(β)]
∂β=E(n)
2+
E(n)
eβE(n) − 1. (1.76)
If we now consider the states with energies that lie in a small range E(n) < E < E(n) + δ,
define g(E) as the density of states:
U =
∫ ∞0
E(n)
eβE(n) − 1g[E(n)]dE(n). (1.77)
16
We recall Equations 1.72 and 1.73, and we discover that this g[E(n)]dE(n) integral is com-
prised of a single octant of a spherical shell; thus, by accounting for two polarizations of light
(an additional factor of two in the number of states) and the g[E(n)]dE(n) = 4πn2dn =
8πL3E2
h3c3dE contribution to the integral is found, and we are able to drop the E(n) notation,
yielding:
U =8πL3
h3c3
∫ ∞0
E3
eβE − 1dE, (1.78)
µ =8πh
c3
∫ ∞0
ν3
ehνkBT − 1
dν, (1.79)
which is an integral we can carry out to find the power radiated per unit area and wave-
length, the Planck blackbody spectrum (in units of Watt per square meter per nanometer):
B(λ, T ) =2hc2
λ51
ehc
λkBT − 1. (1.80)
It is important to note here, that B(λ, T ) is used interchangeably with µ only for a perfect
blackbody. If we wish to extend this to other objects, we must alter the form of µ slightly,
which will be discussed in Section 1.3.2.
Let us now investigate some of the properties of this distribution. We can calculate the
power radiated per unit area as a function of temperature by considering Equation 1.79, and
making the substitution q = hνkBT
, which leads to the integral:
µ = (kBT
h)4
8πh
c3
∫ ∞0
q3
eq − 1dq =
8π5k4B15c3h3
T 4; (1.81)
thus, we recover the power radiated by a blackbody as a function of temperature, the Stefan-
Boltzman law:
µ = σT 4, (1.82)
where σ = 7.566× 10−16 Jm3K4 .
1.3 Kirchoff’s Law
We seek a method to relate the emittance of an object (how well a particular object, or, in our
case, the multilayer structure) emits a given wavelength λ. To achieve this, we will investigate
17
the implications of a simple system and the laws of thermodynamics. It is important to note
that a full proof of Kirchoff’s laws is beyond the scope of this dissertation, and we will
only complete an argument for Kirchoff’s law of thermal radiation here, which speaks to the
link between emittance and absorptance. This law also spoke to the dependence of thermal
radiation on the wavelength of light radiated; however, that portion of the work has since
been overshadowed by Planck’s work on blackbody radiation shown in Equation 1.80.
We consider placing a hot block of material, with emittance εb(λ) and absorptance Ab(λ) in
an extremely-well-insulated container, with emittance εc(λ) = 1 and absorptance Ac(λ) = 1,
a perfect blackbody. The block of material emits radiation. Specifically, we choose to consider
the positive x direction, without loss of generality. At a wavelength, λ0, the radiation which
is emitted over the cross-section dA of the block facing the positive x direction is given by
εb(λ)B(λ, Tb)dA. Similarly, the wall of the container emits radiation which will fall on that
same area dA which is given by εc(λ)B(λ, Tc)dA. The wall of the container absorbs Ac of
incident power and the block absorbs Ab of the incident power, where each 0 < Ac, Ab < 1.
Now, we have that the transfer of energy onto the surface of the block of material:
Q = Ab(λ)εc(λ)B(λ, Tc)dA− εb(λ)B(λ, Tb)dA, (1.83)
which is to say that the container absorbs a fraction Ac(λ) of the incident radiation, and
then re-radiates a portion. Now, If we consider the system in thermodynamic equilibrium,
we know that B(λ, Tb) = B(λ, Tc), the walls are a perfect blackbody, and that the transfer
of energy Q = 0 due to the second law of thermodynamics, which yields:
Ab(λ) = εb(λ). (1.84)
That is, the absorptance of the black and the emittance of the block of material are equal
under the condition of thermal equilibrium. Thus, we have recovered Kirchoff’s law for radi-
ation: In the state of thermal equilibrium, the emittance and absorptance of a given object
18
must be equal. We can now combine Equations 1.80 and 1.84:
µ(λ, θ) = εTotal(λ, θ)2hc2
λ51
ehc
λkBT − 1, (1.85)
which is the power radiated per unit area and unit wavelength by a non-blackbody as a
function of wavelength, angle, and temperature. Here, εTotal(λ, θ) is the averaged emittance
from both the TE and TM polarizations.
1.3.1 Solar Thermophotovoltaics for Power Generation
Traditional solar energy conversion occurs when incident radiation from the Sun falls on
a solar cell, which contains a photovoltaic cell and is converted into usable energy via the
photoelectric effect. This conversion process takes place when an incident photon dislodges
an electron which is orbiting an atom. This now free electron had some transition energy that
is a function of the material used in the photovoltaic cell wafer and is commonly referred
to as the band gap. A maximum conversion efficiency by the solar cell is reached when
incident light on the wafer is comprised only of monochromatic light tuned to the band gap
wavelength λ0.
Figure 1.7 depicts the solar irradiance, the energy incident on the earth’s surface which
is usable by a photovoltaic cell. Solar radiation is broadband. Approximately 42% of solar
radiation lies in the visible; while around 55% is in the infrared. The final 3% is found in the
ultraviolet, with wavelengths less than 400 nm. Because of large spread in incident radiation,
photovoltaic solar cells are extremely inefficient. Photons which have a lower energy than
the bandgap simply cannot excite a transition; while, photons which have a higher energy
than the band gap waste any energy they have in excess of the band gap.
In recent history, there has been much research in the area of these cells. Some research
suggests that employing concentrators to increase the amount of radiation which falls on the
solar cell to both decrease initial investment cost and increase efficiency; however, concen-
trators are extremely bulky and heavy, not suitable for use on spacecraft. Additionally, while
it is cheaper to increase the incident radiation on solar cells with concentrators than it is to
19
FIGURE 1.7: Solar irradiance incident on the earth’s surface in units of watt per metersquared per nanometer. Above the earth’s surface, approximately 1400 watts per metersquared is incident; however, due to the screening effects of the atmosphere, this falls toapproximately 1100 watts per meter squared at sea-level.
simply increase the surface area which is composed of wafers, the concentrators themselves
are still quite expensive.
Other research (and some in-practice solar plants) has developed multi-junction solar cells.
While conventional photovoltaic cells utilize one semiconductor material to convert a small
range of wavelengths efficiently, multi-junction photovoltaic cells use two, three, or even four
different layers of semiconductor materials to efficiently convert a larger portion of the Sun’s
radiation to usable energy.
In 2007, Florescu et al proposed two closely related new designs as seen in Figure 1.8 for
solar cells relying on solar thermophotovoltaics - utilizing the heat from the Sun’s radiation
to cause monochromatic thermal emission from a nanophotonic emitter [8]. A blackbody’s
20
FIGURE 1.8: Schematic of proposed energy conversion device. Radiation from the Sun isincident on an absorbing layer which is in contact with a narrowband emitting layer. Nar-rowband radiation from the emitting layer strikes the solar cell and is converted with highefficiency to usable energy. Here ∆Ω is the solid angle subtended by the Sun at the surfaceof the absorbing layer of the solar cell.
radiant power is given by Equation 1.82 and the efficiency of a Carnot Engine is given by:
η = 1− TcTh, (1.86)
where Th is the temperature of the absorbing layer and Tc is the temperature of the cell. It
is important to note here, that for the purposes of this discussion, we are suppressing the
dependence of emittance on wavelength. In principle, for the absorbing layer, the integral
21
leading to the calculation of Equation 1.82 would need to be numerically evaluated and
depend on ε(λ, θ, φ). However, for this discussion, we will continue to assume that for the
absorbing layer, ε(λ, θ, φ) = 1 and consider the incident radiation on the cell as well as the
energy which is re-radiated by the cell due to spontaneous thermal emission:
Pnet =FsπσT 4
s −FaπσT 4
a , (1.87)
in which each term resembles Equation 1.82 due to the fact that the form factors Faπ
and Fsπ
are the ratios of the solid angle subtended by the Sun and the absorbing cross section to the
total solid angle facing the other. Here, Ts is the surface temperature of the Sun, Ta is the
temperature of the absorbing layer, and σ is the Stefan Boltzmann constant.
Let us now investigate the the normalization factor of π. Equation 1.82 applies to a black-
body with emittance ε = 1 for all wavelengths; here, the Sun is, to a very good approximation,
a blackbody, but is located a large distance from the Earth. This decreases the solid angle
subtended by the Sun and must be accounted for. Similarly, our absorbing layer is not a
perfect blackbody. That is, εa and Aa, the emittance and absorptance, respectively, of the
absorbing layer can be written as εa = Aa < 1. The solid angle subtended by the Sun is
given by:
Fs = Ω =
∫ ∫Surface
r · n dS
r2, (1.88)
where r and n are the unit vectors in the direction of the viewer and to the surface, respec-
tively. For the sum this dot product can be closely approximated as unity. Therefore, we are
left with the cross sectional area of the Sun, and the radius at which the earth orbits the
Sun, leading to Fa = 6.796× 10−5 sr. Following a similar argument and accounting for the
fact that the emittance and absorptance of the absorbing layer are not one, as well as the
fact that the backradiation of the absorbing layer will follow that of a Lambertian object we
can write:
22
Fa =
∫ ∞0
∫ 2π
0
∫ π2
0
εTotal(λ, θ)2hc2
λ51
ehc
λkBT − 1sin(θ) cos(θ) dθ dφ dλ, (1.89)
where, for a one-dimensional multilayer structure, the emittance does not depend on φ. Thus,
we seek to maximize the solar thermophotovoltaic cell efficiency, given by:
ηTPV = (1− FaT4a
FsT 4s
)(1− TcTa
), (1.90)
where the left multiplicand accounts for the efficiency of the Sun-absorber system and the
right multiplicand accounts for the efficiency of the absorber-emitter. If we assume a small
solid angle for absorption such that Fa/Fs is on the order of 100, we note that the theoretical
limit of thermophotovoltaic energy conversion efficiency nears 70% [9].
Thus, we seek to maximize the temperature of the absorbing layer. This can be accom-
plished by a broadband, narrow angle absorbing layer. Our search for such a layer to serve
as a key component of this system is found in Chapter 3.
1.4 Discussion of Optimization Algorithms
Physics in general, and, in particular, computational electricity and magnetism often presents
problems which possess solution spaces which are incredibly vast. For instance, a seemingly
simple 16-layer aperiodic multilayer structure that will be discussed in Chapter 2 has ap-
proximately 2 × 1048 possible solutions, even in the case of discretizing the layer length.
Specifically, in most situations, we choose the smallest layer change to be 1 nm. Thus even
using today’s high-powered computers, finding an optimal solution to this problem by brute
force would likely take years to achieve. Therefore, we must employ optimization algorithms
to quickly and efficiently assess the solution space and produce a solution in a reasonable
period of time. Genetic optimization algorithms [10] are very-well suited for global optimiza-
tion (searching a large portion of a solution space); however, they do not perform efficiently
for local optimization. Therefore, in this dissertation, we employ a genetic optimization al-
gorithm both by itself as well as a hybrid [11] code which couples the genetic algorithm to an
optimization suite packaged by Massachusetts Institute of Technology called NLOPT [12].
23
A genetic algorithm is an optimization procedure which begins with a randomly selected
population of possible solutions. It gradually evolves toward an improved solution by apply-
ing genetic operators which are patterned after the natural selection process. To begin the
calculation, a random selection of values for the genes is created and assigned to a population
of chromosomes. Then, the genetic algorithm iteratively generates a new population by the
crossover, mutation, and selection operators.
The work in this dissertation made use of a microgenetic algorithm, which has been shown
to avoid the problem of premature convergence which is associated with more traditional,
large-population genetic algorithms. The microgenetic algorithm also shows faster conver-
gence to the near-optimal region [13, 14, 15]. A microgenetic algorithm begins with a small
population of random members which evolves and converges after a few generations. Here,
the genetic algorithm selects a new random population, and the evolution process restarts
while keeping the best individual from the previously converged generation.
It is here that in some of the results (particularly in Chapter 4) contained in this document
we make use of the NLOPT optimization package which provides access to extremely efficient
local optimization algorithms. Specifically, we make use of the COBYLA [16] (Constrained
Optimization by Linear Approximations) algorithm contained in NLOPT. This optimization
algorithm maps the local area by examining small changes in the optimization parameters;
then, the algorithm follows this mapping to the local maximum. Once no further improvement
can be found in a small region around the best solution, the new optimized parameters
are returned to the microgenetic algorithm for further optimization. In results presented in
Chapters 2 and 3, however, we do not utilize the NLOPT optimization package.
A tournament selection scheme is employed in this genetic algorithm. Here, we select a
subpopulation of individuals at random from the general population. This subpopulation
competes on the basis of their fitness values. We then choose the individual with the highest
fitness value as the winner and the process is repeated. It has been shown that this selection
24
scheme converges more rapidly and operates more quickly when compared to other competing
schemes [17]. We then select a pair of individuals as the parents, and the crossover operator
creates two offspring. These offspring possess a combination of the chromosomes of their
parents. Here, we use uniform crossover [17, 14] rather than single-point crossover, as it has
been shown that microgenetic algorithm convergence is faster with uniform crossover. For
each generation, we employ an elitist [18] strategy, choosing the best individual from the
previous population to pass on to the next.
25
Chapter 2Optimized Aperiodic Highly DirectionalNarrowband Infrared Emitters
In this chapter, we present optimized aperiodic structures for use as narrowband, highly
directional thermal infrared emitters for both TE and TM polarizations. These aperiodic
multilayer structures designed with alternating layers of silicon and silica on top of a semi-
infinite tungsten substrate exhibit extremely high emittance peaked around the wavelength
at which the structures are optimized. Structures were designed by a genetic optimization
algorithm coupled to a transfer matrix code which computed thermal emittance. First, we
investigate the properties of the genetic-algorithm optimized aperiodic structures and com-
pare them to a previously proposed resonant cavity design. Second, we investigate a structure
optimized to operate at the Wien wavelength corresponding to a near-maximum operating
temperature for the materials used in the aperiodic structure. Finally, we present a structure
that exhibits narrowband and highly directional emittance for both TE and TM polariza-
tions at the frequency of one of the molecular resonances of carbon monoxide (CO); hence,
the design is suitable for use in the emitting portion of a detector of CO via absorption
spectroscopy.1
2.1 Introduction
The thermal emittance of both bulk materials and textured structures has been widely in-
vestigated over recent years. Bulk thermal emittance sources (such as tungsten) possess
incoherent, isotropic, and broadband radiation spectra that vary from material to material;
however, it is well-known that these radiation spectra can be drastically altered by utilizing
textured surfaces [19], multilayer structures, or even three-dimensional constructions [20]
possibly in tandem with a bulk material substrate. From highly directional emitters (anten-
1Part of this chapter previously appeared as C. H. Granier, F. O. Afzal, C. Min, J. P. Dowling, and G. Veronis, “Optimizedaperiodic highly directional narrowband infrared emitters,” J. Opt. Soc. Am. B, 31, 1316, (2014). It is reprinted by permission
of the Optical Society of America (OSA). See the permission letter in the Appendix.
26
nae) to quasi-coherent radiation sources [21, 22] (lasers), as well as solar photovoltaics [23, 9],
these emitters may have many uses due to their emittance spectra altering properties.
There are several approaches that have been used to achieve narrowband, highly direc-
tional thermal emittance. One of these approaches, with one-dimensional photonic crystals,
employed a periodic quarter-wave stack with a half-wavelength resonant cavity as well as
tungsten or silver substrate. This design resulted in the bulk substrate exhibiting direc-
tional, tunable, wavelength-selective emittance [24]. A similar outcome was realized using
periodic one-dimensional metallic photonic crystal slabs [25]. Another approach to achieve
coherent thermal emittance is to use gratings [26, 27, 28] or textured metal surfaces [29].
Other approaches include utilizing metamaterials [30] or shock waves propagating through
a crystal [31, 32] to achieve this end. Finally, narrowband, highly directional transmittance
can be achieved by a photonic heterostructure consisting of two, defective, one-dimensional
photonic crystals [33].
In this chapter, we focus on aperiodic multilayer structures of alternating layers composed
of silicon and silica above a tungsten substrate to produce narrowband, highly directional
thermal emission for both TE and TM polarizations. One-dimensional layered structures
without texturing are preferable to more complex two- and three-dimensional structures
because of the relative ease and low cost of fabrication. Our narrowband, highly tunable in-
frared emitter exhibits highly narrow angular emittance. We achieve this emittance profile by
utilizing a genetic optimization algorithm [14] to select each layer’s thickness independently
so that the structure is completely aperiodic. The choice of the fitness function proved crucial
in obtaining structures with narrowband highly directional emittance. Such a device should
have applications not only as a quasi-coherent radiation source, but also a gas-detection
scheme which will be discussed.
The remainder of this chapter will be organized into three sections. The theory section, (II),
will discuss the computational techniques used. The theory section is followed by the results,
27
(III), which is broken into three subsections. In the first subsection, (III.A) of the results,
we investigate the properties of a genetic-algorithm optimized aperiodic design and compare
them to a previously proposed resonant cavity design. A discussion of an emitter tuned to
the Wien wavelength of a near maximum operating temperature for the structure follows
in the next subsection, (III.B). Lastly, we discuss a structure designed for an application as
the emitting portion of a low-cost carbon monoxide (CO) detector (III.C) and provide our
conclusions in Sec. IV.
2.2 Theory
We envision a structure composed of infinite slabs of material of varying aperiodic thicknesses
as depicted in Figure 2.1. Light is incident from air at an angle θ to the structure. Utilizing the
transfer matrix method [34], we calculate the transmittance, reflectance, and absorbance of
the structure for both TE and TM polarized light.We make use of experimental data for the
wavelength-dependent indexes of refraction, both real and imaginary parts, for silica, silicon,
and tungsten [35] for the calculations done in this chapter. Since the tungsten substrate is
taken to be semi-infinite, the transmittance is identically zero, so that:
ATE/TM(λ, θ) = 1−RTE/TM(λ, θ), (2.1)
where ATE/TM is the TE/TM absorbance, RTE/TM is the TE/TM reflectance, and λ is the
wavelength. While we only calculate absorbance, reflectance, and, in principle, transmittance,
we make use of Kirchhoff’s second law and conservation of energy to equate absorbance
(ATE/TM) and emittance (εTE/TM) under thermal equilibrium [36] .
2.2.1 Fitness
We are interested in finding structures with highly directional and narrowband thermal emit-
tance. We used a genetic optimization algorithm to determine the best structure’s dimensions
for varying numbers of layers at a given wavelength, λ0. The genetic algorithm is an iterative
optimization procedure, which starts with a randomly selected population of potential so-
lutions, and gradually evolves toward improved solutions, via the application of the genetic
28
operators. These genetic operators are patterned after the natural selection process. In the
initialization function, a population of chromosomes is created by random selection of values
for the genes. The genetic algorithm then proceeds to iteratively generate a new population
by the application of selection, crossover, and mutation operators.
More specifically, here we use the microgenetic algorithm. It has been shown that the
microgenetic algorithm avoids premature convergence and shows faster convergence to the
FIGURE 2.1: Schematic of structure optimized by genetic algorithm coupled to the transfermatrix code. Incident light at an angle θ to the normal of the surface of the structure entersthe n-layer alternating structure of silicon and silica above a semi-infinite tungsten substrate.
29
near-optimal region compared to the conventional large-population genetic algorithm for
multidimensional problems [13, 14, 15]. The microgenetic algorithm starts with a small ran-
dom population which evolves and converges after a few generations. At this point, keeping
the best individual from the previously converged generation, a new random population is
chosen, and the evolution process restarts.
We use the tournament selection as the selection scheme in the genetic algorithm. In this
method, a subpopulation of individuals is randomly chosen from the population and made
to compete on the basis of their fitness values. The individual in the subpopulation with
the highest fitness value wins the tournament, and is thus selected. The remaining members
of the entire subpopulation are then put back into the general population, and the process
is repeated. This selection scheme converges more rapidly and has a faster execution time
compared to other competing schemes [17]. Once a pair of individuals is selected as parents,
the basic crossover operator creates two offspring by combining the chromosomes of their
parents. We use uniform crossover rather than single point crossover, as it has been found
that microgenetic algorithm convergence is faster with the uniform crossover [17, 14]. An
elitist strategy [18] is also employed, wherein the best individual from one generation is
passed on to the next generation.
Specifically, we calculated the emittance of each structure as a function of angle for 0 ≤
θ ≤ 90. We then minimized the fitness function, F (λ0),
F (λ0) =
90∫0
εTotal(λ0, θ) dθ, (2.2)
subject to the constraint that εTotal(λ0, θ = 0) ≥ 0.95. Here, εTotal(λ0, θ) = [εTE(λ0, θ) +
εTM(λ0, θ)]/2. That is, we calculated the integral of the emittance over all angles θ at a given
wavelength, λ0, and minimized it subject to the constraint that the emittance at normal
incidence was at least 95%. It is interesting to note that the fitness function that we use
does not impose any constraints on the wavelength dependence of the emittance, since the
30
structures are optimized at a single wavelength; however, simply by minimizing the integral
of the emittance over all angles at a single wavelength [as seen in Figure 2.2, for example],
we also achieve narrowband emittance.
2.3 Results
In this section, we discuss three genetic algorithm optimized structures for use as thermal
emitters. The first structure is optimized to operate at λ0 = 2.357 µm for ease of comparison
to a previously proposed resonant cavity design. The second structure is optimized to operate
at the Wien wavelength corresponding to a near-maximum operating temperature for the
materials used in the structure. Finally, the third structure discussed is optimized to operate
at one of the molecular transition frequencies for CO for possible use in the emitting portion
of an absorption spectroscopy detector.
2.3.1 Aperiodic Emitter
Using the transfer matrix method and genetic algorithm as outlined in Sec. II, we investigated
the properties of the aperiodic one-dimensional structures. We chose λ0 = 2.357 µm as a first
example in the infrared wavelength range. For comparison, we define the angular full width
at half maximum (FWHM) δθn for the n-layer structure, calculating the width about θ = 0
for which the emittance is larger than half of the maximum achieved value as seen in Figure
2.2. We also define the spectral FWHM δλn for the n-layer structure, calculating the width
about the wavelength at which we optimized the structure, λ0, for which the emittance is
larger than half of the maximum achieved value as seen in Figure 2.3.
We consider structures of six, eight, and sixteen layers composed of alternating layers
of silicon and silica. In each case, we minimize the fitness function, F (λ0), [Equation 2.2]
subject to the constraint that the normal incidence emittance εTotal(λ0, θ = 0) ≥ 0.95.
The optimized six-layer structure shows highly directional emittance with δθ6 = 10.8
[Figure 2.2]. The optimized eight-layer structure improves upon this with δθ8 = 7.2. The
optimized sixteen-layer structure improves on the six-layer and eight-layer designs by nearly
31
a factor of four and more than a factor of two, respectively, having δθ16 = 2.7. We also
compare the genetically optimized structures with a previously proposed design consisting
of a periodic quarter-wavelength stack with a half-wavelength resonant cavity over a semi-
infinite tungsten substrate [24], which will heretofore be referred to as the periodic design,
having δθPeriodic = 26, an angular FWHM of more than twice that of the six-layer optimized
structure [Figure 2.2].
FIGURE 2.2: Emittance versus angle of an aperiodic multilayer structure of six, eight, andsixteen alternating layers of silicon and silica over a semi-infinite tungsten substrate. Thestructure is optimized such that the integral of the emittance over all angles for λ0 =2.357µm is minimized subject to the constraint that the emittance at normal incidence is greaterthan 0.95. The layer thicknesses of the optimized structure (in units of microns) are: 1.53,0.46, 0.17, 0.36, 2.32, 2.0 for the six-layer structure, 2.21, 0.45, 1.83, 1.92, 1.18, 1.49, 0.19,2.07 for the eight-layer structure, and 2.25, 1.28, 0.49, 1.95, 2.27, 1.14, 0.85, 2.03, 0.33,1.94, 2.34, 2.28, 0.92, 2.36, 1.27, 1.18 for the sixteen-layer structure. The red dashed linedepicts a four-layer quarter-wave stack of alternating layers of silicon and silica followed bya half-wavelength resonant cavity over a semi-infinite tungsten substrate tuned to λ0 =2.357µm.
32
Moving to the spectral FWHM, we note that the six-layer optimized structure has δλ6 =
8.8 nm [Figure 2.3]. In comparison, the eight-layer optimized structure improves by more
than a factor of two with δλ8 = 3.1 nm. The sixteen-layer structure further improves this
property by a factor of six over the eight-layer structure with a δλ16 = 0.5 nm. Compared to
the genetic-algorithm-optimized aperiodic structures, the periodic structure exhibits much
more broadband emittance, having δλPeriodic = 39.7 nm [Figure 2.3]. These results can be
found for at-a-glance comparison in Table 2.1.
In Figure 2.4 we show the profile of the electric field amplitude, normalized with respect to
the field amplitude of the incident plane wave for the six-layer genetic-algorithm-optimized
aperiodic structure. The structure is excited by a normally incident plane wave at the res-
onant wavelength. We observe a large resonant enhancement of the electric field inside the
FIGURE 2.3: Emittance versus wavelength of the same structures described in Figure 2.2 atnormal incidence.
33
TABLE 2.1: Angular FWHM δθn and spectral FWHM δλn of the structures described inFigure 2.2.
n δθn δλn6 10.8 8.8 nm8 7.2 3.1 nm16 2.7 0.5 nmPeriodic 26 39.7 nm
silica layer adjacent to the tungsten substrate, which leads to greatly enhanced absorption
in the substrate. Thus, one can think of this structure as a silica cavity sandwiched between
a tungsten substrate and a partially reflective aperiodic multilayer stack composed of alter-
nating layers of silicon and silica. We also found that, when the wavelength or the angle of
FIGURE 2.4: Profile of the electric field amplitude, normalized with respect to the fieldamplitude of the incident plane wave for the six-layer genetic-algorithm-optimized aperiodicstructure described in Figure 2.2. The structure is excited by a normally incident plane waveat the resonant wavelength of λ0 = 2.357 µm.
34
incidence are shifted away from the resonance, the field enhancement rapidly decreases. This
is consistent with the narrowband, highly directional emittance of the structure (Figure 2.2).
Overall, the periodic structure exhibited both highly directional and narrowband emit-
tance; however, even the six-layer genetic-algorithm-optimized aperiodic structure improved
upon the directionality and increased the wavelength selectivity of the emittance, with the
eight-layer and sixteen-layer structures further improving these features. As depicted in Fig-
ure 2.2, it is important to note there is a trend of increasing angular selectivity as the number
of layers increases. In addition, we found that there is a decrease in the outlying emittance
in the θ > 60 range as the number of layers increases. In Figure 2.3, a similar trend is seen:
the wavelength selectivity of the emittance of the structure increases with the number of
layers. Even the six-layer aperiodic structure shows vastly increased performance over the
previously seen periodic design. The eight-layer aperiodic structure shows further increased
FIGURE 2.5: Angular emittance of the six-layer aperiodic structure described in Figure 2.2.
35
performance, and the sixteen-layer aperiodic structure exhibits both extreme angular and
spectral selectivity.
We then investigated the off-normal behavior of the emittance for both the genetic-
algorithm-optimized aperiodic structures [Figure 2.5] as well as the periodic structure [Figure
2.6]. Similar trends were seen in the off-normal emittance bands as were seen with the normal.
We looked specifically at the emittance at 30 and 60 and selected the peak wavelength emit-
ted at those angles. There is a trend that as the angle θ increases, the peak of the emittance
shifts to shorter wavelengths. This is due to the fact that the phase accumulation inside each
layer decreases as the angle of incidence increases. This decrease can be compensated for by
decreasing the wavelength [37]. This trend can also be seen in Figure 2.7 where we show the
emittance as a function of wavelength and angle for the six-layer genetic-algorithm-optimized
aperiodic structure. Qualitatively, the periodic structure’s peaks possess larger FWHM for
FIGURE 2.6: Angular emittance for the periodic structure described in Figure 2.2.
36
both wavelength and angular emittance peaks than the genetic-algorithm-optimized ape-
riodic structures. We also note that the genetic algorithm-optimized aperiodic structures’
peaks shift more in angle with decreasing wavelength than do those of the resonant cavity
design. This difference is related to the higher quality factor of the resonances, Q = λ/δλ, of
the aperiodic structures, which leads to higher sensitivity to changes in angle of incidence.
FIGURE 2.7: Emittance as a function of wavelength and angle for the six-layer genetic-algorithm-optimized aperiodic structure described in Figure 2.2.
37
2.3.2 Wien Wavelength
The Planck blackbody spectrum introduces temperature and wavelength dependence into
the spectral radiance B(λ, T ) of a blackbody:
B(λ, T ) =2hc2
λ51
ehc
λkBT−1, (2.3)
where h is Planck’s constant, c is the speed of light, and kB is the Stefan-Boltzman constant.
In principle, the power spectrum µ(λ) for any material or structure with emittance ε(λ) is:
µ(λ) = ε(λ)B(λ, T ). (2.4)
Here, we seek to maximize such emittance while accounting for the melting points of the
materials used. We can calculate the maximum of B as a function of wavelength by setting
∂B/∂λ = 0. This results in the following equation:
xex
ex − 1= 5, (2.5)
where x = hc/λkBT. Once solved, Wien’s displacement law is extracted:
λmax =hc
xkBT, (2.6)
where x is the solution of Equation 2.5, x = 4.96511....
Considering melting points for silicon, silica, and tungsten, silicon has the lowest melting
point of 1690 K. Allowing a small safety margin, we reduce this temperature to 1500 K and
make use of Wien’s displacement law to calculate the appropriate wavelength for maximum
power output to be λ = 1.931 µm. As before, we consider structures of six, eight, and sixteen
layers composed of alternating layers of silicon and silica. In each case, we minimize the
fitness function F (λ0), [Equation(2.2)], subject to the constraint that the normal incidence
emittance εTotal(λ0, θ = 0) ≥ 0.95.
We chose this wavelength and optimized the same six-, eight-, and sixteen-layer aperiodic
structures for angular and wavelength selective emittance here. Again, as we saw in Figure 2.2
38
& Figure 2.3, the same trends appear in Figure 2.8 & Figure 2.9, with the six-layer structure
showing a well-defined peak in both angular and wavelength space, the eight-layer structure
improving upon what the six-layer offered, and the sixteen-layer further increasing angular
and wavelength selectivity. More concretely, the six-layer structure offers a selective emitter,
with a δθ6 = 14.4 and δλ6 = 11.1 nm. The eight-layer optimized structure performs better
by nearly a factor of two with a δθ8 = 8.1 and δλ8 = 3.2 nm. The sixteen-layer optimized
structure outperforms the eight-layer by another factor of two and is a full factor of four
FIGURE 2.8: Emittance versus angle of an aperiodic multilayer structure of six, eight, andsixteen alternating layers of silicon and silica over a semi-infinite tungsten substrate. Thestructure is optimized such that the integral of the emittance for λ0 =1.931 µm over all anglesis minimized subject to the constraint that the emittance at normal incidence is greater than0.95. The layer thicknesses of the optimized structure (in units of microns) are: 1.81, 0.36,0.12, 1.83, 0.43, 1.09 for the six-layer structure, 1.81, 0.29, 0.72, 0.35, 0.72, 1.82, 0.72, 0.26for the eight-layer structure, and 1.55, 0.9, 1.56, 1.63, 1.21, 1.62, 0.39, 1.24, 1.30, 0.34, 0.77,1.87, 1.54, 0.23, 0.24, 0.24 for the sixteen-layer structure.
39
TABLE 2.2: Angular FWHM δθn and spectral FWHM δλn of the structures described inFigure 2.8.
n δθn δλn6 14.4 11.1 nm8 8.1 3.2 nm16 3.6 0.5 nm
better than the six-layer structure with a δθ16 = 3.6 and δλ16 = 0.5 nm. These results can
be found for at-a-glance comparison in Table II.
2.3.3 Aperiodic Structure as the Emitting Portion of a Detector
Carbon Monoxide (CO) is a colorless, odorless, and tasteless gas that is poisonous to hu-
mans. One of the molecular resonances of CO occurs at λ = 4.7 µm [38]. Here, we consider
FIGURE 2.9: Emittance versus wavelength of the same aperiodic structures described inFigure 2.8 at normal incidence.
40
constructing the emitting portion of a low-cost detector for CO utilizing a quasi-coherent
thermal source emitting at the transition frequency.
By heating the designed structure, light is emitted at one of the resonant frequencies of
CO. Via absorption spectroscopy, one may use a detector which searches for dips in the
intensity of the 4.7µm light which would signify an increase of concentration of CO in the
air, thereby setting off the alarm.
With this application in mind, we chose λ0 = 4.7 µm, and we designed optimized struc-
tures. These results are depicted in Figure 2.10 & Figure 2.11. Again, the same trends are
found that increasing the number of layers both increases the angular and wavelength se-
lectivity. At 4.7µm, the selectivity of each the six-, eight-, and sixteen-layer structures was
notably higher than at previously optimized wavelengths.
At this wavelength, the six-layer structure performed a factor of two better than the
structure optimized for λ0 = 1.931 µm with a δθ6 = 8.2 and δλ6 = 4.4 nm. The eight-layer
structure showed marginal improvement over the six-layer with δθ8 = 5.4 and δλ8 = 2.2 nm;
however, the sixteen-layer optimized structure showed more than a factor of three overall
improvement, with a δθ16 = 1.8 and δλ16 = 0.2 nm. These results can be found for at-a-
glance comparison in Table III.
TABLE 2.3: Angular FWHM δθn and spectral FWHM δλn of the structures described inFigure 2.10.
n δθn δλn6 8.2 4.4 nm8 5.4 2.2 nm16 1.8 0.2 nm
2.4 Conclusion
We have designed a series of aperiodic structures that have been optimized to produce
narrow-angular emittance profiles and also display narrow-wavelength emittance. We com-
pared the angular and spectral FWHM of the genetic-algorithm-optimized aperiodic struc-
41
tures to the ones of the resonant cavity design. We have demonstrated that angular selectivity
increases with the number of layers and noticed a strong correlation between angular selec-
tivity and spectral selectivity. We have presented an additional structure that has a use
as a quasi-coherent emitter optimized to operate at the Wien wavelength corresponding to
a near-maximum operating temperature for a silica-silicon structure. Finally, we have out-
lined a low-cost method for constructing the emitting portion of an absorption spectroscopy
detector for CO.
FIGURE 2.10: Emittance versus angle of an aperiodic multilayer structure of six, eight, andsixteen alternating layers of silicon and silica over a semi-infinite tungsten substrate. Thestructure is optimized such that the integral of the emittance for λ =4.7 µm over all anglesis minimized subject to the constraint that the emittance at normal incidence is greater than0.95. The layer thicknesses of the optimized structure (in units of microns) are: 3.77, 0.9,3.76, 4.49, 4.62, 3.16 for the six-layer structure, 3.75, 0.91, 2.40, 4.53, 4.51, 1.28, 3.93,2.94 for the eight-layer structure, and 4.47, 0.88, 0.39, 1.02, 0.35, 0.91, 1.14, 4.00, 2.03,2.39, 1.72, 3.38, 0.34, 0.99, 3.04, 4.15 for the sixteen-layer structure.
42
As final remarks, we have discussed several specific structures that can be tuned to emit
at a wide variety of wavelengths. We note that due to the wavelength dependence of the
dielectric permittivity of the materials used in the design, one cannot simply scale the layer
thicknesses by a factor to scale the operating wavelength by that same factor. In other words,
the optimized designs are not scale invariant, so the optimization process must be carried
out at the new desired wavelength. In all cases, we found that the optimal structures were
unique. In other words, if the stochastic genetic optimization process is repeated, it converges
to the same result. Finally, for structures designed as emitters, there is a question of how
performance will change as temperature is varied. It is known that temperature variations
tend to primarily modify the material losses [39]. We found that, if the material losses
change, we can still use the same genetic algorithm optimization approach to design highly
FIGURE 2.11: Emittance versus wavelength of the same aperiodic structures described inFigure 2.10 at normal incidence.
43
directional, narrowband emitters. Thus, our approach can be used to design structures at
a desired operating temperature. We also found that the effect of layer thickness variations
due to thermal expansion on emittance can be neglected [39].
This research was supported by the National Science Foundation (Award Nos. 1102301,
1263236, 0968895), and a Fund for Innovation in Engineering Research (FIER) grant from
the Louisiana State University College of Engineering. Jonathan P. Dowling wishes to also
acknowledge support from the Air Force Office of Scientific Research and the Army Research
Office.
44
Chapter 3Optimized Aperiodic Multilayer Structuresfor Use as Narrow-Angular Absorbers
In this chapter, we investigate aperiodic multilayer structures for use as narrow-angular
absorbers. The layer thicknesses and materials are optimized using a genetic global opti-
mization algorithm coupled to a transfer matrix code to maximize the angular selectivity in
the absorptance at a single or multiple wavelengths. We first consider structures composed
of alternating layers of tungsten and silicon or silica, and find that it is not possible to
achieve angular selectivity in the absorptance with such structures. We next consider struc-
tures composed of alternating layers of silicon and silica, and show that when optimized
they exhibit high angular selectivity in absorptance. In addition, as the angular selectivity
in absorptance increases, the wavelength range of high angular selectivity also decreases.
Optimizing the material composition of the multilayer structures, in addition to optimizing
the layer thicknesses, leads to marginal improvement in angular selectivity. Finally, we show
that by optimizing the absorptance of the multilayer structures at multiple wavelengths, we
can obtain structures exhibiting almost perfect absorptance at normal incidence and narrow
angular width in absorptance at these wavelengths. Similar to the structures optimized at a
single wavelength, the wavelength range of high angularly selective absorptance is narrow.1
3.1 Introduction
The absorptance properties of materials and structures have been widely investigated in
recent years. Bulk materials, such as tungsten, exhibit broadband and broad-angle absorp-
tion spectra which vary from material to material. However, it has been demonstrated that
the absorption spectra of bulk materials can be drastically changed by using multilayer
structures[40], textured surfaces, or other three dimensional constructions[41, 42]. Several
1Part of this chapter previously appeared as C. H. Granier, F. O. Afzal, S. G. Lorenzo, M. Reyes Jr., J. P. Dowling, and G.Veronis, “Optimized aperiodic multilayer structures for use as narrow-angular absorbers,” J. Appl. Phys. 116 243101 (2014).
It is reprinted by permission of the American Institute of Physics (AIP). See the permission letter in the Appendix.
45
different structures with modified absorptance properties have been investigated, including
metallic gratings[43, 44], metallic and semiconductor photonic crystals[45, 46, 47], periodic
grooves[48], and nano-antennae[49]. The use of metamaterials has also been investigated as a
possible method for tailoring the absorptance characteristics.[30] Developing such structures
that have unusual absorption characteristics in the optical wavelength range is important
for many applications such as photovoltaics,[50, 51, 20] control of thermal radiation,[52, 53]
photodetectors,[54, 55] and chemical sensing[56, 46].
Most related efforts have been focused on developing structures with wide-angle near-
perfect absorption in either a narrow or a broad wavelength range [57, 58, 59, 60, 61, 62,
63, 64, 65, 66]. However, structures which exhibit angular selectivity in absorptance are also
investigated because of their many potential applications [67, 68]. In particular, in solar ther-
mophotovoltaics solar radiation is absorbed by an intermediate absorber, which then emits
thermal radiation towards a solar cell. Solar thermophotovoltaic cells are capable of achieving
theoretical efficiency which far exceeds the Shockley-Queisser limit when a single-junction
cell is directly exposed to sunlight [20, 8]. In order to approach such efficiency, however,
there are very important constraints on the properties of the intermediate absorber. For
efficient solar thermophotovoltaic systems which do not employ sunlight concentration, the
intermediate absorber must exhibit broadband angular selectivity. More specifically, it must
provide near-perfect absorption in a narrow angular range, corresponding to the solid angle
subtended by the Sun, over the entire solar spectrum.[20, 8, 69] Efficient solar thermopho-
tovoltaic systems must have relatively high temperature [69]. From the Earth’s surface, the
solid angle subtended by the Sun is very small. However, a conventional planar surface emits
radiation isotropically in all 2π steradians. Due to this angular mismatch, the planar surface
reaches thermal equilibrium at a much lower temperature than the Sun’s temperature. An-
gular selectivity in the absorption of the intermediate absorber is therefore highly desired in
the case of no sunlight concentration [8, 69], since then we can avoid the loss of energy by
46
the radiation of the intermediate radiator outside of the solid angle subtended by the Sun,
a process which cools the absorber. It was also recently demonstrated that it is possible to
design structures which exhibit broadband angular selectivity in transmission by tailoring
the overlap of the band gaps of multiple photonic crystals, so as to preserve the characteristic
Brewster modes in a broad wavelength range [70].
In this chapter, we investigate one-dimensional multilayer aperiodic structures for use as
narrow-angular absorbers. It has been demonstrated that multilayer structures can tailor the
absorptance spectra of bulk materials.[23] In many cases a multilayer structure may provide
spectra-altering properties similar to that of more complex and harder-to-fabricate two- or
three-dimensional structures. Here, we optimize the layer thicknesses and materials using
a genetic global optimization algorithm coupled to a transfer matrix code to maximize the
angular selectivity in the absorptance at a single or multiple wavelengths. We first consider
structures composed of alternating layers of tungsten and silicon or silica over a tungsten
substrate. We find that, due to the high absorption of tungsten, it is not possible to achieve
angular selectivity in the absorptance with such structures. We next consider structures com-
posed of alternating layers of silicon and silica. Unlike the structures containing tungsten,
the optimized silicon-silica structures exhibit high angular selectivity in absorptance at the
wavelength at which they are optimized. However, as the angular selectivity in absorptance
increases, the wavelength range of high angular selectivity decreases. We then consider opti-
mizing the material composition of the multilayer structures, in addition to optimizing the
layer thicknesses, and find that this approach leads to marginal improvement in angular
selectivity when compared to the silicon-silica structures. Finally, we investigate optimiz-
ing the absorptance of the multilayer structures at multiple wavelengths. We find that with
this approach we can obtain structures exhibiting almost perfect absorptance at normal in-
cidence and narrow angular width in absorptance at these wavelengths. However, similar
47
to the structures optimized at a single wavelength, the wavelength range of high angularly
selective absorptance is narrow.
We previously showed that optimized aperiodic multilayer structures can lead to narrow-
band, highly directional thermal infrared emitters [71]. In this work, structures are designed
to operate in the visible. We note that the properties of the materials used are very differ-
ent in these two wavelength ranges. Silicon is essentially lossless in the infrared, while it is
quite lossy in the visible. As a result, the mechanisms that lead to high angular selectiv-
ity in absorptance are very different. In the infrared designs all the power is absorbed in
the tungsten substrate[71]. In the visible designs presented in this work only a portion of
the power is absorbed in the substrate. In fact, in some of the designs almost no power is
absorbed in the substrate. One can think of such structures as lossy multilayer resonators
which are almost perfectly impedance matched to air. In this work, we also show that it is
not possible to achieve angular selectivity in the absorptance with structures composed of
alternating layers of metals, such as tungsten, and dielectrics due to the high material loss
of metals. In addition, we investigate the effect of optimizing the material composition of
multilayer structures, in addition to optimizing the layer thicknesses. Finally, we show that
with the proposed aperiodic structures it is possible to achieve almost perfect absorptance
at normal incidence and narrow angular width in absorptance at multiple closely-spaced
tunable wavelengths.
The remainder of this manuscript is divided into three sections. The computational tech-
niques used will be discussed in Section II. The results are described in Section III, which
is subdivided into four subsections. The first subsection, (III.A) discusses the results of
genetic-algorithm optimized structures composed of tungsten and silicon as well as tungsten
and silica. Subsection III.B presents the results of genetic-algorithm optimized silicon-silica
structures for use as angular selective absorbers. In Subsection III.C, we investigate opti-
mizing the material composition of the multilayer structures, in addition to optimizing the
48
layer thicknesses. Subsection III.D presents genetic-algorithm-optimized aperiodic structures
composed of alternating layers of silicon and silica with narrow angular absorptance at two
wavelengths (a narrow-angular, bichromatic absorber). Our conclusions are summarized in
Section IV.
3.2 Theory
We model an aperiodic structure composed of infinite slabs of material of varying thicknesses
as depicted in Figure 3.1. Light is incident from air at an angle θ to the structure. Utilizing
the transfer matrix method [34], we calculate the transmittance, reflectance, and absorptance
of the structure for both TE (transverse electric) and TM (transverse magnetic) polarized
light. We make use of experimental data for the wavelength-dependent indexes of refraction,
both real and imaginary parts, for silicon carbide, silica, silicon, and tungsten [35] for the
calculations done in this chapter. These materials are commonly used in applications related
to engineering the absorptance properties of different structures [24, 20], and have high
melting points. In addition, silicon and silica provide a high index contrast which is useful
for engineering the photonic band structure. It is important to note that in the visible range of
wavelengths, both silica and silicon carbide are almost lossless. Since the tungsten substrate
is taken to be semi-infinite, the transmittance is identically zero, so that:
ATE/TM(λ, θ) = 1−RTE/TM(λ, θ), (3.1)
where ATE/TM is the TE/TM absorptance, RTE/TM is the TE/TM reflectance, and λ is the
wavelength.
We are interested in finding structures which are highly absorbing for normally incident
light and also exhibit highly directional absorptance.
We use a genetic optimization algorithm to determine the best structure’s dimensions
for varying numbers of layers at a given wavelength λ0. The implementation of the genetic
algorithm that we use has been realized in-house. A genetic algorithm is an iterative opti-
mization procedure, which starts with a randomly selected population of potential solutions,
49
FIGURE 3.1: Schematic of aperiodic structure optimized by genetic algorithm coupled tothe transfer matrix code. Incident light at an angle θ to the normal of the surface of thestructure enters the n-layer structure comprised of various materials above a semi-infinitetungsten substrate.
and gradually evolves toward improved solutions, via the application of the genetic operators.
These genetic operators are patterned after the natural selection process. In the initializa-
tion function, a population of chromosomes is created by random selection of values for the
genes. The genetic algorithm then proceeds to iteratively generate a new population by the
application of selection, crossover, and mutation operators.
50
More specifically, here we use the microgenetic algorithm. It has been shown that the
microgenetic algorithm avoids premature convergence and shows faster convergence to the
near-optimal region compared to the conventional large-population genetic algorithm for
multidimensional problems [13, 14, 15]. The microgenetic algorithm starts with a small ran-
dom population which evolves and converges after a few generations. At this point, keeping
the best individual from the previously converged generation, a new random population is
chosen, and the evolution process restarts.
We use the tournament selection as the selection scheme in the genetic algorithm. In this
method, a subpopulation of individuals is randomly chosen from the population and made
to compete on the basis of their fitness values. The individual in the subpopulation with
the highest fitness value wins the tournament, and is thus selected. The remaining members
of the entire subpopulation are then put back into the general population, and the process
is repeated. This selection scheme converges more rapidly and has a faster execution time
compared to other competing schemes [72]. Once a pair of individuals is selected as parents,
the basic crossover operator creates two offspring by combining the chromosomes of their
parents. We use uniform crossover rather than single point crossover, as it has been found
that microgenetic algorithm convergence is faster with the uniform crossover [14, 72]. An
elitist strategy is also employed, wherein the best individual from one generation is passed
on to the next generation.
Specifically, for the structures discussed in Subsections III.A, III.B, and III.C, we calculate
the absorptance of each structure as a function of angle for 0 ≤ θ ≤ 90 at wavelength λ0.
We then minimize the fitness function, F (λ0),
F (λ0) =
π2∫
0
ATotal(λ0, θ) dθ, (3.2)
subject to the constraint that ATotal(λ0, θ = 0) ≥ Amin. Here, ATotal(λ0, θ) = [ATE(A0, θ) +
ATM(λ0, θ)]/2. That is, we calculate the integral of the absorptance over all angles θ at a
51
given wavelength, λ0, and minimize it subject to the constraint that the absorptance at
normal incidence is at least Amin.
3.3 Results
In this section, we discuss genetic-algorithm-optimized structures for use as directional ab-
sorbers. Each structure is optimized for angular selectivity in an attempt to find a broadband,
narrow-angle absorber. The structures discussed in Subsections III.A, III.B, and III.C are
optimized at λ0 = 550 nm, roughly the center of the visible range. The structure discussed
in Subsection III.D is optimized to operate as a narrow-angular, bichromatic absorber. It is
optimized at both λ1 = 450 nm and λ2 = 550 nm, which correspond to indigo and green
light, respectively, in the visible range.
3.3.1 Tungsten-Silicon and Tungsten-Silica Structures
Using the transfer matrix method and genetic algorithm as outlined in Sec. II, we investigate
the properties of the aperiodic one-dimensional structures. We choose λ0 = 550 nm at the
center of the visible spectrum as the wavelength at which to optimize angular selectivity of
the absorptance.
First, we consider structures of six, eight, and sixteen layers composed of alternating layers
of tungsten and silicon over a semi-infinite tungsten substrate. In each case, we minimize the
fitness function, F (λ0), [Equation (3.2)] subject to the constraint that the normal incidence
absorptance, ATotal(λ0, θ) ≥ Amin = 0.90. For comparison of the structures, we define the
angular full width at half maximum (FWHM), δθn, for the n-layer structure, calculating
the width about θ = 0 for which the absorptance is larger than half of the maximum
achieved value as seen in Figure 3.2. We also define the spectral FWHM, δλn, for the n-layer
structure, calculating the width about the wavelength at which we optimized the structure,
λ0, for which the absorptance is larger than half of the maximum achieved value as seen in
Figure 3.3. These results for the tungsten-silicon structure are found in Table 3.1.
52
FIGURE 3.2: Absorptance versus angle of an optimized, aperiodic multilayer structure ofsix, eight, and sixteen alternating layers of tungsten and silicon over a semi-infinite tungstensubstrate. The structure is optimized such that the integral of the absorptance over allangles for λ0 = 550nm is minimized subject to the constraint that the absorptance at normalincidence is greater than 90% (Amin = 0.90). The layer thicknesses of the optimized tungsten-silicon structures (in units of nanometers) are: 0, 222, 17, 117, 34, 511 for the six-layerstructure, 0, 357, 17, 51, 18, 48, 30, 313 for the eight-layer structure, and 0, 166, 0, 37,0, 154, 17, 51, 17, 56, 15, 50, 26, 43, 34, 443 for the sixteen-layer structure. It is importantto note that, for some layers, the optimized thickness was found to be zero. We found thatfor the opimized tungsten-silicon structures the top layer, which is adjacent to air, is alwayssilicon.
TABLE 3.1: Angular FWHM δθn and spectral FWHM δλn of the tungsten-silicon structuresdescribed in Figure 3.2.
n δθn δλn6 161.4 70.4 nm8 161.6 42.6 nm16 161.8 45.4 nm
53
FIGURE 3.3: Absorptance versus wavelength of the same structures described in Figure 3.2at normal incidence.
We observe that the genetic-algorithm-optimized tungsten-silicon structures do not exhibit
angular selectivity [Figure 3.2]. In addition, the structures exhibit no wavelength-selectivity
and the absorptance is relatively broadband in nature. The structures exhibit absorptance of
more than 50% in a broad wavelength range [Figure 3.3]. There is very little change in the an-
gular FWHM, δθn, of the respective silicon-tungsten genetic-algorithm-optimized structures
with increasing number of layers (Table 3.1). Overall, the silicon-tungsten material combina-
tion fails to provide angular selectivity and increasing number of layers does not improve the
angular selectivity. This is due to the high absorption of tungsten. In Figure 3.7, we show the
profile of the electric field amplitude, normalized with respect to the field amplitude of the
incident plane wave for the six-layer genetic-algorithm-optimized aperiodic tungsten-silicon
structure. The structure is excited by a normally incident plane wave at the wavelength of
54
FIGURE 3.4: Profile of the electric field amplitude, normalized with respect to the field ampli-tude of the incident plane wave for the six-layer genetic-algorithm-optimized tungsten-siliconaperiodic structure described in Figure 3.2. Note that the thickness of the first tungsten layeris zero; thus, the layer adjacent to air is silicon. The structure is excited by a normally inci-dent plane wave at the wavelength of λ0 = 550 nm. The ratio of the power absorbed insideeach layer to the total power absorbed in the structure was calculated and from left to right,beginning with air, is: 0, .273, .323, .06, .274, .03, .04. That is, ∼27.3% of the power isabsorbed in the first silicon layer adjacent to air; while, ∼4% is absorbed in the tungstensubstrate.
λ0 = 550 nm. We observe that the structure is almost perfectly impedance-matched to air,
since only ∼10% of the incident power is reflected. For this tungsten-silicon structure, the
first two layers provide the majority of the absorption in the structure, contributing ∼27%
and ∼32%, respectively. The second tungsten layer provides ∼27%; while, the substrate
provides less than ∼5 % of the overall absorption of the structure. In the silicon-tungsten
structure, it is obvious from Figure 3.4 that there is no field enhancement provided by the
structure. We therefore conclude that the high absorptance at normal incidence is not asso-
55
FIGURE 3.5: Absorptance versus angle of an optimized aperiodic multilayer structure ofsix, eight, and sixteen alternating layers of tungsten and silica over a semi-infinite tungstensubstrate. The structure is optimized such that the integral of the absorptance over allangles for λ0 = 550nm is minimized subject to the constraint that the absorptance at normalincidence is greater than 90% (Amin = 0.90). The layer thicknesses of the optimized tungsten-silica structures (in units of nanometers) are: 2, 406, 21, 171, 34, 137 for the six-layerstructure, 0, 395, 3, 207, 21, 165, 34, 144 for the eight-layer structure, and 0, 413, 9,192, 22, 168, 22, 155, 26, 293, 0, 392, 17, 163, 13, 517 for the sixteen-layer structure. It isimportant to note that, for some layers, the optimized thickness was found to be zero. Wealso found that for the optimized tungsten-silicon structures the top layer, which is adjacentto air, is in the case of the eight and sixteen layer structure, silicon.
ciated with any strong resonances. This is also supported by the broad angle and wideband
absorptance depicted in Figures 3.2 and 3.3.
We also consider structures of six, eight, and sixteen layers composed of alternating layers
of tungsten and silica. Again, we minimize the fitness function, F (λ0), [Equation (3.2)]
subject to the constraint that the normal incidence absorptance, ATotal(λ0, θ) ≥ Amin = 0.90.
56
FIGURE 3.6: Absorptance versus wavelength of the same structures described in Figure 3.5at normal incidence.
TABLE 3.2: Angular FWHM δθn and spectral FWHM δλn of the tungsten-silica structuresdescribed in Figure 3.5.
n δθn δλn6 74.6 88.4 nm8 72.9 82.6 nm16 68.4 71.0 nm
Similar to the tungsten-silicon structures, the tungsten-silica structures do not exhibit high
angular selectivity in absorptance. As depicted in Figure 3.5, the tungsten-silica structures
all achieve a peak of 90% absorptance at normal incidence and exhibit reduced angular
FWHM, δθn, (Table 3.2) when compared to the tungsten-silicon structures. However, the
tungsten-silica structures maintain high absorptance for large angles of incidence [Figure 3.5].
We note that the tungsten-silica structures possess more broadband absorptance, having a
larger spectral FWHM (δλn) compared to the tungsten-silicon structures (Table 3.2). In
57
FIGURE 3.7: Profile of the electric field amplitude, normalized with respect to the fieldamplitude of the incident plane wave for the six-layer genetic-algorithm-optimized tungsten-silica aperiodic structure described in Figure 3.5. The structure is excited by a normallyincident plane wave at the wavelength of λ0 = 550 nm. The ratio of the power absorbedinside each layer to the total power absorbed in the structure was calculated and from left toright, beginning with air, is: 0, .541, 0, .302, 0, .13, 0, .03. That is, ∼54.1% of the power isabsorbed in the first tungsten layer adjacent to air; while, ∼3% is absorbed in the tungstensubstrate.
Figure 3.7, we show the profile of the electric field amplitude, normalized with respect to
the field amplitude of the incident plane wave for the six-layer genetic-algorithm-optimized
aperiodic tungsten-silica structure. As before, the structure is excited by a normally incident
plane wave at the wavelength of λ0 = 550 nm. In the tungsten-silica structure, ∼85% of
the absorption occurs in the first two tungsten layers. The substrate provides less than
∼5% of the overall absorptance in the structure. As with the tungsten-silicon structure, the
absorptance mechanism is non-resonant and is the result of impedance matching.
Overall, due to the high absorption of tungsten, we found that it is not possible to achieve
angular selectivity with structures using tungsten. In both tungsten-based structures, the
58
high absorptance at normal incidence is not associated with resonant field enhancement. In
addition, in both cases the structures were impedance-matched to air in a broad wavelength
range.
3.3.2 Silica-Silicon Structures
FIGURE 3.8: Absorptance versus angle of an optimized aperiodic multilayer structure of four,six, eight, and sixteen alternating layers of silicon and silica over a semi-infinite tungstensubstrate. The structure is optimized such that the integral of the absorptance over allangles for λ0 = 550nm is minimized subject to the constraint that the absorptance at normalincidence is greater than 0.95 (Amin = 0.95). The absorptance of bulk tungsten is depicted inblack for reference. The layer thicknesses of the optimized silicon-silica structures (in unitsof nanometers) are: 28, 530, 33, 301 for the four-layer structure, 30, 508, 32, 340, 34, 88for the six-layer structure, 25, 110, 34, 530, 34, 92, 33, 80 for the eight-layer structure,and 12, 172, 28, 120, 34, 496, 34, 503, 0 (silicon), 550, 25, 321, 27, 106, 36, 81 for thesixteen-layer structure. It is important to note that in some cases the optimized thicknesswas found to be zero.
We consider structures of four, six, eight, and sixteen layers composed of alternating layers
of silicon and silica over a semi-infinite tungsten substrate. In each case, we minimize the
59
FIGURE 3.9: Absorptance versus wavelength of the six-layer structure described in Figure 3.8at normal incidence. Again, absorptance of bulk tungsten is depicted in black for reference.
fitness function, F (λ0), [Equation (3.2)] subject to the constraint that the normal incidence
absorptance at λ0 = 550 nm, ATotal(λ0, θ) ≥ Amin = 0.95. In doing so, we seek structures
which provide narrow angular absorptance profiles.
TABLE 3.3: Angular FWHM δθn and spectral FWHM δλn of the silicon-silica structuresdescribed in Figure 3.8.
n δθn δλn4 28.8 12.8 nm6 20.8 6.6 nm8 13.3 2.6 nm16 6.9 0.8 nm
The genetic-algorithm-optimized, silicon-silica structures exhibit angular selectivity at
λ = 550 nm, the wavelength at which the structures are optimized. As before, the FWHM
comparison of the optimized four, six, eight, and sixteen layer structures can be found in
60
FIGURE 3.10: Absorptance versus wavelength of the four, six, eight, and sixteen layer struc-tures described in Figure 3.8 and zoomed in on the wavelength for which the structure wasoptimized. Absorptance of bulk tungsten is depicted in black for reference.
Table 3.3. All of the silicon-silica structures exhibit angular selectivity, with each structure
peaking at 95% absorptance at normal incidence and decreasing rapidly as θ increases [Fig-
ure 3.8]. We also note that there is a trend of increasing angular selectivity as the number
of layers increases. For comparison, the absorptance of bulk tungsten at this wavelength
is also shown. Figure 3.9 shows the spectral response at normal incidence of the six-layer
genetic-algorithm-optimized, silicon-silica structure as well as bulk-tungsten’s response. The
six-layer structure provides a narrow band response centered around the wavelength at which
it is optimized with additional resonances at other wavelengths. The eight and sixteen layer
structures’ spectral response at normal incidence is similar. As the number of layers in the
optimized structure increases, the number of resonances in the λ = 400 nm to λ = 700 nm
wavelength range also increases. Figure 3.10 shows the absorptance of the four, six, eight,
61
FIGURE 3.11: Absorptance as a function of wavelength and angle for the eight-layer, genetic-algorithm-optimized, aperiodic structure described in Figure 3.8.
and sixteen layer structures near λ = 550 nm, the wavelength at which the structures are
optimized. We observe that there is a trend of decreasing spectral resonance width with
increasing number of layers in the optimized structure. Figure 3.11 shows a color plot of the
absorptance as a function of both angle and wavelength for the eight-layer, genetic-algorithm-
optimized structure. We observe that the structure exhibits angular selectivity around the
wavelength at which it was optimized, but its angular occurrence shifts with wavelength.
More specifically, as the wavelength decreases, the peak of the absorptance shifts to larger
62
angles. Overall, with increasing number of layers, we observe an increase in angular selectivity
and a decrease in the wavelength range of high absorptance at normal incidence.
FIGURE 3.12: Profile of the electric field amplitude, normalized with respect to the fieldamplitude of the incident plane wave for the six-layer genetic-algorithm-optimized aperiodicstructure described in Figure 3.8. The structure is excited by a normally incident plane waveat the wavelength of λ0 = 550 nm. The ratio of the power absorbed inside each layer to thetotal power absorbed in the structure was calculated and from left to right, beginning withair, is: 0, .062, 0, .06, 0, .107, 0, .77.
In Figure 3.12 and Figure 3.13 we show the profile of the electric field amplitude, nor-
malized with respect to the field amplitude of the incident plane wave for both the six-layer
and eight-layer genetic-algorithm-optimized, silicon-silica aperiodic structures. The structure
is excited by a normally incident plane wave at λ0 = 550 nm, the wavelength at which the
structures are optimized. In the six-layer structure [Figure 3.12], we observe a large, resonant
enhancement of the electric field toward the center of the structure. As a result, the major-
ity (∼77%) of the power is absorbed in the tungsten substrate. In the eight-layer structure
63
FIGURE 3.13: Profile of the electric field amplitude, normalized with respect to the field am-plitude of the incident plane wave for the eight-layer genetic-algorithm-optimized aperiodicstructure described in Figure 3.8. The structure is excited by a normally incident plane waveat the wavelength of λ0 = 550 nm. The ratio of the power absorbed inside each layer to thetotal power absorbed in the structure was calculated and from left to right, beginning withair, is: 0, .048, 0, .29, 0, .318, 0, .046, 0, .3. That is, ∼5% of the power is absorbed in thefirst silicon layer adjacent to air; while, ∼30% is absorbed in the tungsten substrate.
[Figure 3.13], the enhancement of the field is approximately twice the enhancement of the
six-layer structure; however, the majority of the power is absorbed outside of the tungsten
substrate. Less than one-third of the overall power is absorbed in the tungsten substrate. The
two largest contributors to power absorbed for the eight-layer structure are the second and
third silicon layers with the second silicon layer absorbing ∼29% and the third silicon layer
absorbing ∼32% of the total power absorbed by the structure. Overall, we found that with
increasing number of layers, the fraction of the power absorbed in the substrate decreases.
We also found that, when the wavelength or the angle of incidence are shifted away from the
resonance, the field enhancement rapidly decreases. This is consistent with the narrowband,
64
highly directional absorptance of the structure (Figure 3.8). Additionally, it is important to
note that as the number of layers increases, the field enhancement drastically increases.
In short, unlike the structures containing tungsten, optimized, multilayer, silicon-silica
structures exhibit high angular selectivity for absorptance at the wavelength at which they
are optimized. However, we found that, for such optimized multilayer silicon-silica structures,
as the angular FWHM, δθn, decreases, the spectral FWHM, δλn, also decreases. In other
words, as the angular selectivity in absorptance increases, the wavelength range of high
angular selectivity decreases.
3.3.3 Multiple Material Structures
We also investigated optimizing the material composition of the multilayer structures in
addition to optimizing the layer thicknesses. As before, we minimize the fitness function,
F (λ0), [Equation (3.2)] subject to the constraint that the normal incidence absorptance at
λ0 = 550 nm, ATotal(λ0, θ) ≥ Amin = 0.95. In doing so, we seek structures which provide
narrow angular absorptance profile. More specifically, for each layer, the material was chosen
among silicon, silica, tungsten, and silicon carbide. In each case, the genetic optimization
algorithm searched for the optimal material composition as well as the optimal thickness of
each layer. We considered structures consisting of four, six, eight, and sixteen layers. In all
cases except for the four-layer structure, silicon carbide was found to be the optimal material
for the first layer adjacent to air. In addition, none of the optimized multilayer structures
contained tungsten. The fact that the optimal material composition of the multilayer struc-
ture does not contain tungsten is consistent with our conclusion in Subsection III.A that it
is not possible to achieve angular selectivity in the absorptance with structures containing
highly lossy materials such as tungsten.
Even though the optimized material composition was found to be, in general, different
from the one of the silicon-silica multilayer structures, the improvement in angular selectivity
with respect to silicon-silica structures with the same number of layers was marginal. More
65
specifically, for a given number of layers n, the angular FWHM, δθn of the structures with
optimized material composition was no more than 2% decreased with respect to the silicon-
silica multilayer structures with the same number of layers.
3.3.4 Bichromatic Absorber
FIGURE 3.14: Absorptance versus angle of an aperiodic multilayer structure of eight alter-nating layers of silicon and silica over a semi-infinite tungsten substrate. The structure isoptimized such that the integral of the sum of the absorptances at the wavelengths at whichthe structure was optimized over all angles, θ, is minimized subject to the constraint thatthe absorptance at normal incidence was at least Amin = 0.95 at all wavelengths considered.Simultaneously, we constrained the angle-average difference between the absorptances at thetwo wavelengths at which the structures were optimized to be less than 0.01. The layer thick-nesses of the optimized silicon-silica structure (in units of nanometers) for the eight-layerstructure are: 14, 300, 69, 248, 21, 127, 245, 463.
As described in Subsection III.B, optimized silicon-silica multilayer structures exhibit high
angular selectivity in absorptance when optimized at a single wavelength in the visible wave-
length range. However, as the angular selectivity increases, wavelength selectivity also in-
66
FIGURE 3.15: Absorptance versus wavelength of the same structure described in Figure 3.14at normal incidence.
creases. In an attempt to increase the wavelength range of high angular selectivity in absorp-
tance, we investigate optimizing the angular selectivity of multilayer structures at multiple
wavelengths.
More specifically, we calculate the absorptance of each structure as a function of angle for
0 ≤ θ ≤ 90 at two wavelengths, λ1 = 450 nm and λ2 = 550 nm in the visible wavelength
range. Then we minimize the fitness function, F1(λ1, λ2),
F (λ1, λ2) =
π2∫
0
[ATotal(λ1, θ) + ATotal(λ2, θ)] dθ, (3.3)
subject to the constraints ATotal(λi, θ = 0) ≥ Amin = 0.95, i = 1, 2. Here, ATotal(λi, θ) =
[ATE(λi, θ) + ATM(λi, θ)]/2.
67
FIGURE 3.16: Profile of the electric field amplitude, normalized with respect to the field am-plitude of the incident plane wave for the eight-layer, genetic-algorithm-optimized, aperiodicstructure described in Figure 3.14. The structure is excited by a normally incident plane waveof λ1 = 450 nm (blue) and λ2 = 550 nm (black, dashed). The ratio of the power absorbedinside each layer to the total power absorbed in the structure was calculated and from leftto right, beginning with air, is: 0, .217, 0, .519, 0, .025, 0, .196, 0, .04 for λ1 = 450 nm, and0, .053, 0, .596, 0, .021, 0, .155, 0, .17 for λ2 = 550 nm. That is, for λ1 = 450 nm, ∼22%of the power is absorbed in the first silicon layer adjacent to air; while, ∼4% is absorbed inthe tungsten substrate.
That is, we calculate the integral of the sum of the absorptances at the wavelengths at
which the structure is optimized over all angles, θ, and minimize it subject to the constraint
that the absorptance at normal incidence is at least Amin = 0.95 at all wavelengths consid-
ered. However, we found that with this optimization process, the absorptance response of
the optimal structures was not unique. This is due to the fact that structures with identical
fitness, F1(λ1, λ2) may exhibit different angular selectivity at the wavelengths at which the
structures are optimized. As an example, we consider two distinct structures. One of the
68
structures exhibits higher angular selectivity at wavelength λ1, [δθ(λ1) < δθ(λ2)]; the other
one exhibits higher angular selectivity at λ2 [δθ(λ1) > δθ(λ2)], while both have the same
fitness, F1(λ1, λ2).
To address this, we added one additional constraint in the optimization process:
F2(λ1, λ2) =2
π
π2∫
0
|ATotal(λ1, θ)− ATotal(λ2, θ)| dθ ≤ ε. (3.4)
That is, we constrained the angle-average difference between the absorptances at the two
wavelengths at which the structures were optimized to be less than ε. Here, we chose ε = 0.01.
In other words, the average difference in the angular absorptance profiles has to be less than
1%.
Here, we consider a structure composed of eight alternating layers of silicon and silica above
a thick tungsten substrate. We found that at least eight layers are required in the silicon-silica
structure in order to achieve high angular selectivity at the two wavelengths at which the
absorptance of the structure was optimized. The structure provides relatively narrow-angular
absorptance features at both of the wavelengths at which it was optimized, λ1 = 450 nm and
λ2 = 550 nm. In fact, the angular absorptance spectra for the two wavelengths are nearly
identical [Figure 3.14]. The absorptance of the optimized structure at normal incidence as
a function of wavelength is shown in Figure 3.15. The structure absorbs the incident light
almost completely at the two wavelengths at which it was optimized; however, similar to the
structures optimized at a single wavelength (Subsection III.B), the wavelength range of high
angularly selective absorptance is narrow (Table 3.4).
TABLE 3.4: Angular FWHM δθn and spectral FWHM δλn of the structures described inFigure 3.14.
Wavelength δθ δλλ1 = 450 nm 36.7 8.4 nmλ2 = 550 nm 36.6 10.6 nm
69
In Figure 3.16, we show the profile of the electric field amplitude, normalized with re-
spect to the field amplitude of the incident plane wave for the eight-layer, genetic-algorithm-
optimized, aperiodic structure. The structure is excited by a normally incident plane wave
at either λ1 = 450 nm or λ2 = 550 nm. While the wavelengths at which the structure was
optimized have nearly identical angular spectra, the two absorptance mechanisms within
the structure are different. The λ1 = 450 nm light is absorbed mainly in the first two sili-
con layers, which account for over ∼70% of the structure’s absorbed power. The tungsten
substrate accounts for less than ∼5% of the overall power absorbed at λ1 = 450 nm. The
λ2 = 550 nm light is absorbed mostly in the second silicon layer, which provides ∼60% of
the overall absorption at this wavelength; the final silicon layer before the tungsten substrate
and the substrate itself account for an additional ∼30% of the structure’s absorbed power.
3.4 Conclusion
We investigated one-dimensional aperiodic multilayer structures for use as narrow-angular
absorbers. We focused on structures which are highly absorbing for normally incident light
and also exhibit highly directional absorptance. We optimized the layer thicknesses and
materials using a genetic global optimization algorithm coupled to a transfer matrix code to
maximize the angular selectivity in the absorptance at a single or multiple wavelengths.
We first considered structures composed of alternating layers of tungsten and silicon or
silica over a tungsten substrate. We found that, due to the high absorption of tungsten,
it is not possible to achieve angular selectivity in the absorptance with such structures.
In addition, the absorptance of such tungsten-based structures is relatively broadband in
nature, and is not associated with any strong resonances.
We next considered structures composed of alternating layers of silicon and silica. Unlike
the tungsten-based multilayer structures, the optimized silicon-silica structures exhibit a
resonance and high angular selectivity in absorptance at the wavelength at which they are
optimized. The resonance is associated with large field enhancement in the structures. We
70
also found that there is a trend of increasing angular selectivity as the number of layers
increases. However, as the angular selectivity in absorptance increases, the wavelength range
of high angular selectivity decreases.
We then considered optimizing the material composition of the multilayer structures, in
addition to optimizing the layer thicknesses. For each layer, the material was chosen among
silicon, silica, tungsten, and silicon carbide. Even though the optimized material composition
was found to be, in general, different from the one of the silicon-silica multilayer structures,
the improvement in angular selectivity with respect to silicon-silica structures with the same
number of layers was marginal.
Finally, we investigated optimizing the absorptance of the multilayer structures at multi-
ple wavelengths. We found that this approach leads to structures exhibiting almost perfect
absorptance at normal incidence and narrow angular width in absorptance at these wave-
lengths. However, similar to the structures optimized at a single wavelength, the wavelength
range of high angularly selective absorptance is narrow.
As final remarks, we note that due to the wavelength dependence of the dielectric permit-
tivity of the materials utilized in the design, one cannot simply scale the layer thicknesses by
a factor to scale the operating wavelength by that same factor. In other words, the optimized
designs are not scale invariant, so the optimization process must be carried out at the new
desired wavelength. It is also important to note that the optimized designs for the tungsten-
silica and tungsten-silicon multilayer structures are non-unique. In other words, there are
multiple designs which achieve the minimum fitness. On the other hand, the optimized de-
signs for the silicon-silica multilayer structures are unique. In addition, we note that it is
possible to design silicon-silica multilayer structures which exhibit high angular selectivity in
absorptance at visible wavelengths without the tungsten substrate. In these structures the
light is absorbed in the silicon layers, since silicon is a lossy material in the visible.
71
It should also be noted that at the wavelength of λ = 550 nm the real part of the di-
electric permittivity of tungsten is positive. Thus, at this wavelength tungsten behaves as
an extremely lossy dielectric rather than as a plasmonic metal. The very high absorption of
tungsten broadens all resonances both in angle and wavelength. In fact, we found that, if
one assumes that tungsten is lossless [by setting εtungsten = Re(εtungsten) and neglecting the
imaginary part Im(εtungsten)], then it is possible to design multilayer structures containing
this artificially lossless tungsten which exhibit high angular selectivity in absorptance.
In addition, we note that angular selectivity in absorptance can also be achieved by a
structure consisting of a quarter-wave one-dimensional photonic crystal placed over a perfect
mirror with an absorbing material included in the defect layer formed by the photonic crystal
and the mirror. In the one-dimensional photonic crystal structure the number of layers in
the photonic crystal has to be adjusted in order to get unity absorptance at the resonance
[24]. However, once the number of layers of the photonic crystal is chosen, the angular width
of the absorptance cannot be tuned. Compared to such a structure, our proposed structures
have many more degrees of freedom. Thus, it is possible to simultaneously obtain perfect
absorptance on resonance and also tune the angular width of the absorptance. Our proposed
aperiodic multilayer structures can achieve much narrower angular width than the one-
dimensional photonic crystal structures. In addition, with the proposed aperiodic structures
it is possible to achieve perfect absorptance at normal incidence and narrow angular width in
absorptance at multiple closely-spaced tunable wavelengths. Such an absorptance response
cannot be achieved with the one-dimensional photonic crystal structures.
This research was supported by the National Science Foundation (Award Nos. 1102301,
1263236, 1403105), and a Fund for Innovation in Engineering Research (FIER) grant from
the Louisiana State University College of Engineering. Jonathan P. Dowling wishes to also
acknowledge support from the Air Force Office of Scientific Research and the Army Research
Office.
72
Chapter 4How to Build a Better Lightbulb —Exploiting Optimization Algorithms toDecrease Incandescent Lightbulb PowerConsumption by Nearly 50 %
In this chapter, we present optimized aperiodic structures for use as broadband, broad-angle
thermal emitters which are capable of increasing the emittance by nearly a factor of two over
the visible wavelength range when compared to bulk tungsten. These aperiodic multilayer
structures designed with alternating layers of tungsten and air or tungsten and silicon carbide
on top of a tungsten substrate exhibit high emittance in a broad wavelength range peaked
around the center of the visible. We optimize the layer thicknesses using a hybrid optimization
algorithm coupled to a transfer matrix code to maximize the power emitted in the visible
wavelength range in the normal direction. We investigate the properties of these structures
for use as lightbulb filaments, and compare their performance with conventional lightbulbs.
We find that these structures greatly enhance the emittance over the visible wavelength
range, while also increasing the overall efficiency of the bulb. These structures could lead to
a decrease in incandescent lightbulb power consumption by nearly 50 %.
4.1 Introduction
Conventional incandescent lightbulbs (ILB) are composed of a tungsten filament inside a
bulb which is filled with inert gas. These devices are less than 10 percent efficient [73, 74]
due to the fact that most of the applied power is radiated as heat and not as visible light.
There has been much recent work in an attempt to improve the efficiency of household and
commercial lighting. Compact fluorescent lamps (CFLs), the curled bulbs, improve efficiency
slightly, but have issues with slow turn on time, higher cost, degradation of performance over
CFL lifetime, and most significantly, these bulbs contain mercury and pose an environmental
hazard if broken or disposed in a landfill. The most efficient lighting commercially available
73
is the light-emitting diode (LED) which possesses efficiencies between 25 and 35 percent.
LEDs, however, are extremely expensive (due to fabrication costs) which puts them out of
reach of users in emergent countries. Hence, incandescent bulbs still account for the vast
majority of home lighting in developing countries; while, in the United States, the ILB has
been phased out to be replaced with CFL or LED lighting.
In this chapter, we propose one-dimensional multilayer aperiodic structures for use as
incandescent lightbulb filaments. It has been shown that multilayer structures can tailor
the emittance spectra of bulk materials and may provide properties similar to that of more
complex and harder-to-fabricate two- or three-dimensional structures [40, 19, 21, 42, 23, 25,
26, 45, 46, 48, 49, 57, 58, 59, 60]. Here, we optimize the layer thicknesses using a hybrid
optimization algorithm coupled to a transfer matrix code to maximize the power emitted
in the visible wavelength range in the normal direction. First, we consider a tungsten-air
structure above a tungsten substrate as a proof of concept. We next consider a tungsten and
silicon-carbide structure for possible use as a filament in an incandescent lightbulb.
4.2 Theory
We model a structure composed of infinite slabs of material of varying aperiodic thicknesses
as depicted in Figure 4.1. Utilizing the transfer matrix method, [34] we calculate the trans-
mittance, reflectance, and absorptance of the structure for both TE and TM polarized light.
Light is incident from air at an angle θ to the structure. We make use of experimental data
for the wavelength-dependent indexes of refraction, both real and imaginary parts, for silicon
carbide and tungsten [35] in all calculations done in this chapter. We choose silicon carbide
and tungsten due to their high melting point, which is necessary for use in a lightbulb fila-
ment application. Since the tungsten substrate is taken to be semi-infinite, the transmittance
is identically zero, so that:
ATE/TM(λ, θ) = 1−RTE/TM(λ, θ), (4.1)
where ATE/TM is the TE/TM absorptance, RTE/TM is the TE/TM reflectance, and λ is the
74
Tungsten...
d1
d2
d3
d4
θ
θ = 0Ο
FIGURE 4.1: Schematic of the structure optimized. It consists of alternating layers of tungstenand air or tungsten and silicon carbide above a semi-infinite tungsten substrate.
wavelength. While we only calculate absorptance, reflectance, and, in principle, transmit-
tance, we make use of Kirchhoff’s second law and conservation of energy to equate absorp-
tance (ATE/TM) and emittance (εTE/TM) under thermal equilibrium [36] .
More specifically, the power radiated per unit area and wavelength is given by the Planck
blackbody spectrum (in units of Watt per square meter per nanometer):B(λ, T ) = 2hc2
λ5(e
hcλkBT−
1)−1, where h is Planck’s constant, c is the speed of light, λ is the wavelength, T is the tem-
perature, and kB is the Boltzmann constant. We define the normalized power radiated per
unit area and unit wavelength by a given structure, µ(λ, θ) as the ratio of the power emitted
75
per unit area and wavelength by the structure to the maximum emitted by a blackbody
µ(λ, θ) =εTotal(λ, θ)B(λ, T = 2700K)
maxλ
[B(λ, T = 2700K)]. (4.2)
Here, εTotal(λ, θ) = [εTE(λ, θ) + εTM(λ, θ)]/2. We choose an operating temperature of 2700 K,
below the melting point of silicon carbide and tungsten and near the operating temperature
of modern incandescent lightbulbs [75].
Conventional incandescent lightbulbs are made by drawing tungsten into a thin wire and
running a current through the wire. This process, via the resistance of the wire, heats the
wire to a temperature between 2500 K and 3000 K and produces light over broad wavelength
range. Due to the operating temperature of the filament as well as the emittance spectra of
bulk tungsten, this process is inefficient and produces both a large amount of waste heat and
light which is not in the visible wavelength range.
In a traditional incandescent lightbulb, in excess of 90% of the power which enters the bulb
is released as heat. Additionally, of the approximately 10% of the power which is released
as radiation, only a small percentage of the light is emitted in the visible wavelength range
(Figure 4.2). We define the efficiency of a given structure as the ratio of the normalized
power emitted in the visible wavelength range (λ1 = 400 nm to λ2 = 700 nm) to the total
power emitted
η =
λ2∫λ1
µ(λ, θ = 0) dλ
∞∫0
µ(λ, θ = 0) dλ
. (4.3)
Using Equation (4.3), we find that for a bulk tungsten filament operating at T = 2700K,
only ∼ 11.55% of the power radiated lies in the visible wavelength range and is therefore
useful for indoor lighting. The combination of these effects leads to an extremely low efficiency
for conventional incandescent tungsten filament lightbulbs.
Another factor which contributes to the overall low efficiency of conventional tungsten
filaments is the fact that tungsten is highly reflective over the visible wavelength range.
76
FIGURE 4.2: Normalized power emitted per unit area and unit wavelength, at normal inci-dence, µ(λ, θ), as a function of wavelength for the optimized structures compared to botha perfect blackbody at the same temperature as well as a conventional bulk-tungsten fila-ment. The solid black line depicts a perfect blackbody at 2700 K. The solid green line showsthe performance of an ideal filament operating at 2700 K. The dashed gray line shows theperformance of a bulk tungsten lightbulb filament. The solid orange and dashed blue linesshow the performance of the optimized TSiC and TA structures, respectively. The layerthicknesses of the optimized structures, in units of nanometers, beginning with the tungstenlayer bordering air are: 3.4, 90, 163, 500 for the TA structure and 0, 39, 66, 319 for theTSiC structure.
Tungsten is a relatively poor emitter, with an average emittance over the visible spectrum of
approximately 50%. We define the enhancement factor Q of a given structure as the ratio of
the total power emitted in the visible by the structure at T = 2700 K to the power emitted
by a bulk tungsten filament heated to the same temperature:
Q =
λ2∫λ1
µ(λ, θ = 0) dλ
λ2∫λ1
µBulk(λ, θ = 0) dλ
. (4.4)
77
We are interested in finding structures which exhibit a plateau-like emittance over the
visible wavelength range. That is, we seek a structure which possesses ε = 1.0 in the visible
wavelength range (λ1 ≤ λ ≤ λ2) and ε = 0.0 elsewhere (corresponding to the solid green
line in Figure 4.2) in an effort to maximize the power emitted by the structure in the visible
while maximizing its efficiency.
We use a hybrid optimization method consisting of a micro-genetic global optimization al-
gorithm [14, 13, 15, 17, 18, 76] coupled to a local optimization algorithm [12, 16] to determine
the best dimensions for four-layer structures. The genetic algorithm is an iterative optimiza-
tion procedure which starts with a randomly selected population of potential solutions and
evolves toward improved solutions; once the population converges, the local optimization
algorithm finds the local optimum. The process retains the best structure found and is iter-
atively repeated.
Specifically, we first maximize the fitness function F1 where
F1 =
λ2∫λ1
µ(λ, θ = 0) dλ. (4.5)
This process is carried out to enhance the emittance in the visible when compared to that
of bulk tungsten.
We then seek to maximize the efficiency of the structure [Equation (4.3)] without signif-
icantly decreasing the enhancement of the power emitted in the visible compared to bulk
tungsten [Equation (4.4)]. Due to the operating temperature of the structure, T = 2700 K,
any radiation occurring at λ ≤ 400nm can be treated as negligible. Thus, we minimize the
normalized power emitted by the structure for λ ≥ 700 nm, F2,
F2 =
∞∫λ2
µ(λ, θ = 0) dλ, (4.6)
subject to the constraint that the fitness function F1 for the structure is at least 98 % of the
previously calculated maximum F1.
78
4.3 Results
We consider structures of four layers composed of alternating layers of tungsten and air
(henceforth referred to as the TA structures) or alternating layers of tungsten and silicon
carbide (henceforth referred to as the TSiC structures). We optimize the emittance of the
structures using the process outlined above [Equations (4.5) and (4.6)]. Both of the opti-
mized structures show drastically improved emittance (by nearly a factor of two) over the
visible wavelength range when compared to bulk tungsten (Figure 4.2). Thus, these struc-
tures could lead to a decrease in incandescent lightbulb power consumption by nearly 50
%. More specifically, a lightbulb consuming 30 W of power using our design would radiate
approximately the same energy in the visible as a conventional incandescent bulb which
consumed 60 W of power. A detailed comparison of the optimized structures presented here
and current lightbulb technology can be found in Figure 4.2. The emittance of the optimized
structures approaches the one of a perfect blackbody. By comparing the bulk tungsten’s
(Figures 4.3(a) & 4.3(b) and 4.6(a) & 4.6(b)) emittance and normalized power emitted per
unit area and unit wavelength µ to that of both TA (Figures 4.4(a) & 4.4(b) and 4.6(c) &
4.6(d)) and TSiC structures (Figures 4.5(a) & 4.5(b) and 4.6(e) & 4.6(f)), we observe that
the optimized structures provide greatly enhanced emittance over the visible wavelength
range when compared to bulk tungsten. As a result, the optimized structures also greatly
enhance the power emitted by the structures in the visible. A detailed comparison of the
enhancement factor Q and the efficiency η of the different structures can be found in Table
4.1.
TABLE 4.1: Comparison of the enhancement, Q, and the efficiency, η, using bulk tungstenas a reference point, of the optimized structures described in Figure 4.2..
Structure Q η
Tungsten 1.0 11.55%W-Air 1.927 13.62%W-SiC 1.854 13.77%
79
FIGURE 4.3: (a, b) Emittance and normalized power emitted per unit area and unit wave-length µ(λ, θ) of bulk tungsten as a function of wavelength and angle.The layer thicknessesof the optimized structures are given in Figure 4.2.
80
FIGURE 4.4: (a, b) Emittance and normalized power emitted per unit area and unit wave-length µ(λ, θ) for the optimized TA structure as a function of wavelength and angle. Thelayer thicknesses of the optimized structures are given in Figure 4.2.
81
FIGURE 4.5: (a, b) Emittance and normalized power emitted per unit area and unit wave-length µ(λ, θ) for the optimized TSiC structure as a function of wavelength and angle. Thelayer thicknesses of the optimized structures are given in Figure 4.2.
In Figures 4.7 & 4.8, we show the profile of the electric field amplitude normalized with
respect to the field amplitude of the incident plane wave for the optimized TA and TSiC
82
FIGURE 4.6: (a, b) Emittance and normalized power emitted per unit area and unit wave-length µ(λ, θ) of bulk tungsten as a function of wavelength and angle. (c, d) Emittance andnormalized power emitted per unit area and unit wavelength µ(λ, θ) for the optimized TAstructure as a function of wavelength and angle. (e, f) Emittance and normalized power emit-ted per unit area and unit wavelength µ(λ, θ) for the optimized TSiC structure as a functionof wavelength and angle. The layer thicknesses of the optimized structures are given in Figure4.2.
structures, respectively. Each structure is excited by a normally incident plane wave at the
center of the visible spectrum, λ = 550 nm. We observe that both structures are nearly
perfectly impedance-matched to air, as only ∼ 10% of the incident power is reflected. In
addition, there is no field enhancement in the structures. We therefore conclude that the
high emittance is not associated with any strong resonances. This is consistent with the
broad-angle and wideband emittance of the structures (Figure 4.6).
We found that in the optimized TA structure the 500 nm thick air layer [Figure 4.7 is
not critical, since in a structure without this layer the emittance in the normal direction is
smaller by no more than ∼ 0.5%. Thus, the optimized TA structure consists essentially of a
3.4 nm thick tungsten layer separated from the tungsten substrate by a 90 nm thick air gap.
One can think of such a structure as a Fabry-Perot [77, 78] resonator as evidenced by the
83
FIGURE 4.7: Profile of the electric field amplitude, normalized with respect to the fieldamplitude of the incident plane wave for the TA abd TSiC optimized structures described inFigure 4.2. The structures are excited by a normally incident plane wave at the wavelengthof λ0 = 550 nm, near the center of the visible wavelength range. For the TA structure, theratio of the power absorbed inside each layer to the total power absorbed in the structurewas calculated and from left to right, beginning with tungsten, is: 0.79, 0, 0.21, 0, 0. Thatis, ∼79% of the power is absorbed in the first tungsten layer adjacent to air.
response of the structure’s emittance to changing the width of the first two layers [Figure
4.9(a) & 4.9(b)]. Even though, as expected, the optimum air gap thickness to maximize
the absorption in the 3.4 nm thick tungsten layer is wavelength dependent, we found that
a structure with a 90 nm thick air gap leads to at least 73.5% absorption in the tungsten
layer in the whole visible wavelength range. The mechanism which leads to broadband high
emittance in the optimized TSiC structure is similar. In this case the optimized three layer
structure above the tungsten substrate [Figure 4.7] leads to greatly enhanced absorption in
the 66 nm thick tungsten layer in the whole visible wavelength range.
84
FIGURE 4.8: Profile of the electric field amplitude, normalized with respect to the fieldamplitude of the incident plane wave for the TA abd TSiC optimized structures described inFigure 4.2. The structures are excited by a normally incident plane wave at the wavelengthof λ0 = 550 nm, near the center of the visible wavelength range. For the TSiC aperiodicstructure, the thickness of the first tungsten layer is zero; thus, the layer adjacent to air issilicon carbide. The ratio of the power absorbed inside each layer to the total power absorbedin the structure from left to right, beginning with SiC, is: 0, .99, 0, 0.01. That is, ∼99%of the power is absorbed in the first tungsten layer
TABLE 4.2: Comparison of the wattage required to produce the same lumens as a bulktungsten incandescent bulb as well as several common commercial bulbs [79].
Bulb Type Watts Required LumensWatt
LumensW Incandescent 60 15 900Halogen 43 17.4 750CFL Halogen 15 53 800CFL 13 63.5 825Smart LED 13 61.5 800W-Air 31.1 28.9 900W-SiC 32.4 27.8 900
85
FIGURE 4.9: Here, we plot the emittance at normal incidence for λ = 550 nm of the overalloptimized Tungsten-Air structure while altering individual layer thicknesses of the first andsecond layer. (a) Emittance of the Tungsten-Air structure as a function of thickness of thefirst Tungsten layer bordering air on the left hand side of the structure as shown in Figure4.7. (b) Emittance of the Tungsten-Air structure as a function of thickness of the first Airlayer inside the structure shown in Figure 4.7.
86
To implement our proposed lighting scheme we propose manufacturing sheets of the ma-
terial inside of a transparent chamber filled with an inert gas (or evacuated) with current
running through the tungsten substrate to provide the heating needed for thermal emission.
To replace an incandescent light bulb we propose simply taking the usual solid tungsten
filament, which is a cylinder with a ∼ 5 µm diameter and coating it with the multi-layered
structure proposed here. This can be done by vapor deposition or molecular beam epitaxy.
Given that the layers are on the scale of nanometers, the curvature of the filament is negli-
gible; therefore, the planar 1D calculation still holds quantitatively. To reduce evaporation
effects, one might wish to reduce the surface area to volume ratio and go to 2D or 3D struc-
tures (where an air-tungsten structure is feasible to fabricate via, for example, ion etching or
a modified silicon layering process) [41]. More work needs to be done to model and optimize
such structures, but our results in this work indicate qualitatively, via the John-Wang model,
that similar emissivity (or better) is possible. In the John-Wang model, one assumes a 3D
structure that has a perfectly spherical Brillouin zone that is the same for both the TE and
TM modes. In such a model the structure is effectively 1D and our results here should hold
qualitatively [80, 81] .
This research was supported by the National Science Foundation (Award Nos. 1102301,
1254934, 1263236, 0968895), and a Fund for Innovation in Engineering Research (FIER)
grant from the Louisiana State University College of Engineering. Jonathan P. Dowling
wishes to also acknowledge support from the Air Force Office of Scientific Research and the
Army Research Office.
87
Chapter 5Conclusions
In Chapter 1 of this dissertation, we began with a brief introduction to nanophotonics. We
provided a discussion of the theory of computational electricity and magnetism, and, specif-
ically, calculating the absorptance and transmittance of a multilayer structure using the
transfer-matrix method. We discussed the principles which motivated the design and anal-
ysis of multilayer structures. Specifically, we reviewed thermal emission, thermodynamics,
Planck’s blackbody distribution, Kirchoff’s law, and resulting applications in solar ther-
mophotovoltaics. In particular, a detailed discussion of the efficiency and possible implica-
tions of realizing a solar thermophotovolatic device were discussed. We also presented a brief
discussion of the computer-based optimization used. Specifically, we use an in-house genetic
algorithm which may be coupled to a local optimization algorithm for increased optimization
of structure parameters.
Chapter 2 discussed aperiodic multilayer structures which were optimized by a genetic
algorithm to provide both narrowband and narrow-angle thermal emission for selected wave-
lengths in the infrared wavelength range. This chapter provided a detailed comparison of
the performance of aperiodic multilayer structures with the more widely researched periodic
multilayer structures and analyzed the physical mechanisms which helped to achieve this
emission profile via an investigation of the electromagnetic field profile. Chapter 2 also in-
vestigated the use of aperiodic multilayer structures optimized for narrow-angle and narrow
band emission at the Wein Wavelength for a near-maximum operating temperature for sili-
con and silica. Finally, Chapter 2 discussed the application of an aperiodic genetic algorithm
optimized emitter for use as a carbon monoxide detector in an absorption spectroscopy de-
88
vice. The work presented in this chapter could have implications in creating inexpensive,
easy-to-fabricate, tunable infrared thermal emitters.
In Chapter 3, we focused on the development of aperiodic multilayer structures for use as
narrow-angle absorbers utilizing a genetic optimization algorithm in the visible wavelength
range. First, we examined the properties of two tungsten based emitters, one which paired
tungsten with silicon and another which paired tungsten with silica. We found that struc-
tures did not provide narrow-angle absorptance profiles due to the fact that non-resonant
impedance matching was the cause of these structure’s high absorptance around λ = 550nm.
We then examined the properties of aperiodic multilayer structures containing alternating
layers of silicon and silica and found that these structures provided both a narrow angle and
narrow band absorptance profile. We also investigated the properties of structures found
when the genetic algorithm was allowed to choose the layer composition and found no siz-
able improvement over the silicon-silica structured when compared on an equal-layer basis.
Finally, we showed that structures with near-perfect absorptance at two wavelengths for
normally incident light can be achieved using these methods.
Chapter 4 presented optimized aperiodic structures which could be used as broadband
and broad-angle thermal emitters. By using a hybrid optimization algorithm coupled to a
transfer-matrix code, we maximized the power emitted in the visible wavelength range in the
normal direction and then, while holding that constant, maximized the efficiency of these
structures as thermal emitters in the visible wavelength range. We investigated structures
consisting of alternating layers of tungsten and air and tungsten and silicon carbide, re-
spectively. We found that both optimized structures could increase emittance in the visible
wavelength range by nearly a factor of two when compared to bulk tungsten. By investigat-
ing the electromagnetic field profiles, we present the mechanisms present in these optimized
structures which allow them to possess properties which could lead to a decrease in incan-
descent lightbulb power consumption by nearly 50%.
89
Bibliography
[1] S. C. Hagness A. Taflove, editor. Computational Electrodynamics: The Finite-DifferenceTime-Domain Method. Artech House, Inc, 3rd edition, 2005.
[2] J. Joannopoulos, R. Meade, and J. Winn. Photonic Crystals: Molding the Flow of Light.Princeton University Press, 1st edition, 1995.
[3] J. D. Jackson. Classical Electrodynamics. John Wiley and Sons, Inc., 3rd edition, 1999.
[4] A. Zangwill, editor. Modern Electrodynamics. Cambridge University Press, 1st edition,2013.
[5] D. J. Griffiths, editor. Introduction to Electrodynamics. Prentice Hall, 3rd edition, 1999.
[6] D. V. Schroeder. An Introduction to Thermal Physics. Addison Wesley Longman, 1stedition, 2000.
[7] M. Kardar, editor. Statistical Physics of Particles. Cambridge University Press, 3rdedition, 2007.
[8] M. Florescu, H. Lee, I. Puscasu, M. Pralle, L. Florescu, D. Z. Ting, and J. P. Dowling.Improving solar cell efficiency using photonic band-gap materials. Sol. Energy Mater.Sol. Cells, 91:1599, 2007.
[9] M. Florescu, H. Lee, I. Puscasu, M. Pralle, D. Ting L. Florescu, and J. P. Dowling.Improving solar cell efficiency using photonic band-gap materials. Sol. Energy Mater.Sol. Cells, 91:1599, 2007.
[10] D. H. Werner R. L. Haupt, editor. Genetic Algorithms in Electromagnetics. Wiley-IEEEPress, 1st edition, 2007.
[11] E. Michielssen Y. Rahmat-Samii, editor. Electromagnetic Optimization by Genetic Al-gorithms. John Wiley and Sons, Inc., 1st edition, 1999.
[12] S. G. Johnson. The nlopt nonlinear-optimization package, 2008.
[13] D. E. Goldberg and K. Deb. Fundamentals of Statistical and Thermal Physics. MorganKaufmann, 1965.
[14] K. Krishnakumar. Micro-genetic algorithms for stationary and non-stationary functionoptimization. Proc. SPIE, 1196:289, 1989.
[15] K. Deb and S. Agrawal. Understanding interactions among genetic algorithm param-eters. In Fundamentals of Statistical and Thermal Physics, pages 268–269. MorganKaufmann, 1999.
90
[16] M. J. D. Powell. A direct search optimization method that models the objective andconstraint functions by linear interpolation. In S. Gomez and J. P. Hennart, editors,Advances in Optimization and Numerical Analysis, pages 51–67. Kluwer Academic: Dor-drecht, 1994.
[17] D. E. Goldberg, K. Deb, and J. H. Clark. Genetic algorithms, noise, and the sizing ofpopulations. Complex Systems, 6:333–362, 1991.
[18] J. M. Johnson and Y. Rahmat-Samii. Genetic algorithms in electromagnetics. Proc.Int. Symp. IEEE Antennas Propagation, 6:1480–1483, 1996.
[19] M. Ghebrebrhan, P. Bermel, Y. X. Yeng, I. Celanovic, M. Soljacic, and J. D. Joannopou-los. Tailoring thermal emission via q matching of photonic crystal resonances. Phys.Rev. A, 83:033810, 2011.
[20] E. Rephaeli and S. Fan. Absorber and emitter for solar thermo-photovoltaic systems toachieve efficiency exceeding the shockley-queisser limit. Opt. Express, 17:15145, 2009.
[21] J. J. Greffet, K. Joulain R. Carminati, S. Mainguy J. P. Mulet, and Y. Chen. Resonanttransmission through a metallic film due to coupled modes. Nature (London), 416:61,2002.
[22] D. Chan, M. Soljacic, and J. D. Joannopoulos. Thermal emission and design in 2d-periodic metallic photonic crystal slabs. Opt. Express, 14:8785, 2006.
[23] J. G. Fleming, S. Y. Lin, I. El-Kady, R. Biswas, and K.M. Ho. All-metallic three-dimensional photonic crystals with a large infrared bandgap. Nature (London), 417:52,2002.
[24] I. Celanovic, D. Perreault, and J. Kassakian. Resonant-cavity enhanced thermal emis-sion. Phys. Rev. B, 72:075127, 2005.
[25] D. L. C. Chan, M. Soljacic, and J. D. Joannopoulos. Thermal emission and design inone-dimensional periodic metallic photonic crystal slabs. Phys. Rev. E, 74:016609, 2006.
[26] C. Arnold, F. Marquier, F. Pardo M. Garin, S. Collin, N. Bardou, J. L. Pelouard, andJ. J. Greffet. Coherent thermal infrared emission by two-dimensional silicon carbidegratings. Phys. Rev. B, 86:035316, 2012.
[27] M. Laroche, C. Arnold, F. Marquier, R. Carminati, J. Greffet, S. Collin, N. Bardou,and J. Pelouard. Highly directional radiation generated by a tungsten thermal source.Opt. Lett., 30:2623, 2005.
[28] S. Han and D. Norris. Beaming thermal emission from hot metallic bull’s eyes. Opt.Express, 18:4829, 2010.
[29] G. Biener, N. Dahan, A. Niv, V. Kleiner, and E. Hasman. Highly coherent theremalemission obtained by plasmonic bandgap structures. Appl. Phys. Lett., 92:081913, 2008.
91
[30] N. Mattiucci, G. D’Aguanno, A. Alu, C. Argryopoulos, J. Foreman, and M. J. Bloemer.Taming the thermal emissivity of metals: a metamaterial approach. Appl. Phys. Lett.,100:201109, 2012.
[31] E. J. Reed, M. Soljacic, and J. D. Joannopoulos. Maxwell equation simulations ofcoherent optical photon emission from shock waves in crystals. Phys. Rev. E, 75:056611,2007.
[32] E. J. Reed, M. Soljacic, R. Gee, and J. D. Joannopoulos. Molecular dynamics simula-tions of coherent optical photon emission from shock waves in crystals. Phys. Rev. B,75:174302, 2007.
[33] G. Liang, P. Han, and H. Wang. Narrow frequency and sharp angular defect mode inone-dimensional photonic crystals from a photonic heterostructure. Opt. Lett., 29:192,2004.
[34] C. M. Cornelius and J. P. Dowling. Modification of planck blackbody radiation byphotonic band-gap structures. Phys. Rev. A, 59:4736, 1999.
[35] D. R. Lide. CRC Handbook of Chemistry and Physics. CRC Press, 2007.
[36] E. U. Condon. Fundamentals of Statistical and Thermal Physics. McGraw-Hill, 1965.
[37] N. Tessler, S. Burns, H. Becker, and R. H. Friend. Omnidirectional emission from top-emitting organic light-emitting devices with microstructured cavity. Appl. Phys. Lett.,70:556, 1997.
[38] A. L. Smith. The Coblentz Society Desk Book of Infrared Spectra. The Coblentz Society,1982.
[39] Y. X. Yeng, M. Ghebrebrhan, P. Bermel, W. R. Chan, J. D. Joannopoulos, M. Soljacic,and I. Celanovic. Enabling high-temperature nanophotonics for energy applications.PNAS, 109:2280, 2012.
[40] W. Wang, Y. Cui, Y. He, Y. Hao, Y. Lin, X. Tian, T. Ji, and S. He. Efficient multibandabsorber based on one-dimensional periodic metal-dielectric photonic crystal with areflective substrate. Opt. Lett., 39:331, 2014.
[41] S. Y. Lin, J. Moreno, and J. G. Fleming. Three-dimensional photonic-crystal emitterfor thermal photovoltaic power generation. App. Phys. Lett, 83:380, 2003.
[42] E. Rephaeli and S. Fan. Tungsten black absorber for solar light with wide angularoperation range. Appl. Phys. Lett., 92:211107, 2008.
[43] N. Mattiucci, G. D’Aguanno, A. Alu, C. Argyropoulos, J. V. Foreman, and M. J. Bloe-mer. Taming the thermal emissivity of metals: A metamaterial approach. Appl. Phys.Lett., 100:201109, 2009.
[44] J.-J. Greffet, R. Carminati, K. Joulain, J.-P. Mulet, S. Mainguy, and Y. Chen. Coherentemission of light by thermal sources. Nature, 416:61, 2002.
92
[45] Y. X. Yeng, J. B. Chou, V. Rinnerbauer, Y. Shen, S.-G. Kim, J. D. Joannopoulos, andM. Soljacic andI. Celanovic. Global optimization of omnidirectional wavelength selec-tive emitters/absorbers based on dielectric-filled anti-reflection coated two-dimensionalmetallic photonic crystals. Opt. Express, 22:21711, 2014.
[46] G. Veronis, R. W. Dutton, and S. Fan. Metallic photonic crystals with strong broadbandabsorption at optical frequencies over wide angular range. J. Appl. Phys., 97:093104,2005.
[47] G. C. R. Devarapu and S. Foteinopoulou. Mid-ir near-perfect absorption with a sicphotonic crystal with angle-controlled polarization selectivity. Opt. Express, 20:13040,2012.
[48] T. Søndergaard and S. I. Bozhevolnyi. Theoretical analysis of plasmonic black gold:periodic arrays of ultra-sharp grooves. New J. Phys., 15:013034, 2013.
[49] J. A. Schuller, T. Taubner, and M. L. Brongersma. Optical antenna thermal emitters.Nature Photon., 3:658, 2009.
[50] O. Stenzel, A. Stendal, K. Voigtsberger, and C. von Borczyskowski. Enhancement ofthe photovoltaic conversion efficiency of copper phthalocyanine thin film devices byincorporation of metal clusters. Sol. Energy Mater. Sol. Cells, 37:337, 1995.
[51] M. Westphalen, U. Kreibig, J. Rostalski, and H. Luth andD. Meissner. Metal clusterenhanced organic solar cells. Sol. Energy Mater. Sol. Cells, 61:97, 2000.
[52] S. Y. Lin, J. Moreno, and J. G. Fleming. Three-dimensional photonic-crystal emitterfor thermal photovoltaic power generation. Appl. Phys. Lett., 83:380, 2003.
[53] S. Y. Lin, J. G. Fleming, Z. Y. Li, I. El-Kady, R. Biswas, and K. M. Ho. Origin ofabsorption enhancement in a tungsten, three-dimensional photonic crystal. J. Opt. Soc.Am. B, 20:1538, 2003.
[54] A. K. Sharma, S. H. Zaidi, P. C. Logofatu, and S. R. J. Brueck. Optical and electricalproperties of nanostructured metal-silicon-metal photodetectors. J. Opt. Soc. Am. B,38:1651, 2002.
[55] H. Huang, Y. Huang, X. Wang, Q. Wang, and X. Ren. Long wavelength resonant cavityphotodetector based on inp/air-gap bragg reflectors. IEEE Photonics Technol. Lett.,16:245, 2004.
[56] A. Tittl, P. Mai, R. Taubert, D. Dregely, N. Liu, and H. Giessen. Palladium-based plas-monic perfect absorber in the visible wavelength range and its application to hydrogensensing. Nano Lett., 11:4366, 2011.
[57] N. Liu, M. Mesch, T. Weiss, M. Hentschel, and H. Giessen. Infrared perfect absorberand its application as plasmonic sensor. Nano Lett., 10:2342, 2010.
93
[58] M. Diem, T. Koschny, and C. M. Soukoulis. Wide-angle perfect absorber/thermal emit-ter in the terahertz regime. Phys. Rev. B, 79:033101, 2009.
[59] J. Hendrickson, J. Guo, B. Zhang, W. Buchwald, and R. Soref. Wideband perfect lightabsorber at midwave infrared using multiplexed metal structures. Opt. Lett., 37:371,2012.
[60] C. Hu, Z. Zhao, X. Chen, and X. Luo. Realizing near-perfect absorption at visiblefrequencies. Opt. Express, 17:11039, 2009.
[61] T. V. Teperik, F. J. Garcia de Abajo, A. G. Borisov, M. Abdelsalam, P. N. Bartlett,Y. Sugawara, and J. J. Baumberg. Omnidirectional absorption in nanostructured metalsurfaces. Opt. Express, 2:299, 2008.
[62] K. Aydin, V. E. Ferry, R. M. Briggs, and H. A. Atwater. Broadband polarization-independent resonant light absorption using ultrathin plasmonic super absorbers. Nat.Commun., 2:517, 2011.
[63] M. K. Hedayati, M. Javaherirahim, B. Mozooni, R. Abdelaziz, A. Tavassolizadeh,V. S. K. Chakravadhanula, V. Zaporojtchenko, T. Strunkus, F. Faupel, and M. Elbahri.Design of a perfect black absorber at visible frequencies using plasmonic metamaterials.Adv. Mater., 23:5410, 2011.
[64] Y. Cui, K. H. Fung, H. Ma J. Xu, Y. Jin, S. He, and N. X. Fang. Ultrabroadband lightabsorption by a sawtooth anisotropic metamaterial slab. Nano Lett., 12:1443, 2012.
[65] T. Søndergaard, S. M. Novikov, T. Holmgaard, R. L. Eriksen, J. Beermann, Z. Han,K. Pedersen, and S. I. Bozhevolnyi. Plasmonic black gold by adiabatic nanofocusingand absorption of light in ultra-sharp convex grooves. Nat. Commun., 3:969, 2012.
[66] G. C. R. Devarapu and S. Foteinopoulou. Broadband mid-ir superabsorption withaperiodic polaritonic photonic crystals. J. Europ. Opt. Soc. Rap. Public., 9:14012, 2014.
[67] C. Argyropoulos, K. Q. Le, G. D’Aguanno N. Mattiucci and, and A. Alu. Broadbandabsorbers and selective emitters based on plasmonic brewster metasurfaces. Phys. Rev.B, 87:205112, 2013.
[68] Y. Shen, D. Ye, Z. Wang, L. Wang, I. Celanovic, L. Ran, J. D. Joannopoulos, andM. Soljacic. Metamaterial broadband angular selectivity. Phys. Rev. B, 90:125422,2014.
[69] P. Bermel, M. Ghebrebrhan, M. Harradon, Y. X. Yeng, I. Celanovic, J. D. Joannopoulos,and M. Soljacic. Tailoring photonic metamaterial resonances for thermal radiation.Nanoscale Res. Lett., 6:549, 2011.
[70] Y. Shen, D. Ye, I. Celanovic, S. G. Johnson, J. D. Joannopoulos, and M. Soljacic.Optical broadband angular selectivity. Science, 343:1499, 2014.
94
[71] C. H. Granier, F. O. Afzal, C. Min, J. P. Dowling, and G. Veronis. Optimized aperiodichighly directional narrowband infrared emitters. J. Opt. Soc. Am. B, 31:1316, 2014.
[72] D. E. Goldberg, K. Deb, and J. H. Clark. Genetic algorithms, noise, and the sizing ofpopulations. Complex Systems, 6:333–362, 1991.
[73] J. Toedt, D. Koza, and K. V. Cleef-Toedt. Chemical Composition of Everyday Products.Greenwood Press, 2005.
[74] J. M. Berd. Environmental Chemistry in Society. CRC Press, 2nd edition, 2014.
[75] L. A. Bloomfield. How Things Work: The Physics of Everyday Life. John Wiley &Sons, Inc., 4 edition, 2009.
[76] C. H. Granier, F. O. Afzal, Jr. S. G. Lorenzo an M. Reyes, J. P. Dowling, and G. Veronis.Optimized aperiodic multilayer structures for use as narrow-angular absorbers. J. Appl.Phys, 116:243101, 2014.
[77] A. Yariv and P. Yeh, editors. Photonics: Optical Electronics in Modern Communications.Oxford University Press, 6th edition, 2007.
[78] B. E. A. Saleh and M. C. Teich, editors. Fundamentals of Photonics. John Wiley andSons, Inc., 1st edition, 1991.
[79] General Electric. Find the right general electric light bulbs, 2015.
[80] S. John and J. Wang. Quantum electrodynamics near a photonic band gap: Photonbound states and dressed atoms. Phys. Rev. Lett., 64:2418, 1990.
[81] S. John and J. Wang. Quantum optics of localized light in a photonic band gap. Phys.Rev. B, 43:12772, 1991.
95
Appendix: Permission to Use Publications
The following are the copyright policies of the Journal of Optical Society of America Band American Institute of Physics (AIP) from:
• https://www.osapublishing.org/josab/submit/review/copyright permissions.cfm#posting
Author Posting Policy: Transfer of copyright does not prevent an author from subsequentlyreproducing his or her article. OSA’s Copyright Transfer Agreement gives authors the rightto publish the Author Accepted version of the article or chapter in a compilation of theauthor’s own works or reproduce the article for teaching purposes on a short-term basis.
• http://publishing.aip.org/authors/copyright-reuse
Q: May I include my AIP article in my thesis or dissertation?AIP permits authors to include their published articles in a thesis or dissertation. It is
understood that the thesis or dissertation may be published in print and/or electronic formand offered for sale, as well as included in a university’s repository. Formal permission fromAIP is not needed. If the university requires written permission, however, we are happy tosupply it.
96
Vita
Christopher H. Granier was born in 1987 in Baton Rouge, Louisiana. He majored in physics
with a mathematics minor at Louisiana State University and graduated in May, 2010. He
continued his education by entering graduate school at LSU in June 2010 and received his
Masters of Science in Physics in May 2014. He is currently a candidate for the degree of
Doctor of Philosophy in Physics, which is to be awarded in August, 2015.
97
Top Related